[go: up one dir, main page]

CN111079828A - Method, device and storage medium for image clustering based on neural network - Google Patents

Method, device and storage medium for image clustering based on neural network Download PDF

Info

Publication number
CN111079828A
CN111079828A CN201911286423.3A CN201911286423A CN111079828A CN 111079828 A CN111079828 A CN 111079828A CN 201911286423 A CN201911286423 A CN 201911286423A CN 111079828 A CN111079828 A CN 111079828A
Authority
CN
China
Prior art keywords
image
database
images
determining
clustering
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Granted
Application number
CN201911286423.3A
Other languages
Chinese (zh)
Other versions
CN111079828B (en
Inventor
贠瑞峰
刘粉香
彭翔
张炎红
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Shenqi Digital Co ltd
Original Assignee
Smart Shenzhou Beijing Technology Co Ltd
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Smart Shenzhou Beijing Technology Co Ltd filed Critical Smart Shenzhou Beijing Technology Co Ltd
Priority to CN201911286423.3A priority Critical patent/CN111079828B/en
Publication of CN111079828A publication Critical patent/CN111079828A/en
Application granted granted Critical
Publication of CN111079828B publication Critical patent/CN111079828B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/23—Clustering techniques
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/50—Information retrieval; Database structures therefor; File system structures therefor of still image data
    • G06F16/51—Indexing; Data structures therefor; Storage structures
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/50—Information retrieval; Database structures therefor; File system structures therefor of still image data
    • G06F16/58—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually
    • G06F16/583—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually using metadata automatically derived from the content
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/22—Matching criteria, e.g. proximity measures
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02D—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN INFORMATION AND COMMUNICATION TECHNOLOGIES [ICT], I.E. INFORMATION AND COMMUNICATION TECHNOLOGIES AIMING AT THE REDUCTION OF THEIR OWN ENERGY USE
    • Y02D10/00—Energy efficient computing, e.g. low power processors, power management or thermal management

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Data Mining & Analysis (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Artificial Intelligence (AREA)
  • Evolutionary Computation (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Evolutionary Biology (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Software Systems (AREA)
  • Databases & Information Systems (AREA)
  • Library & Information Science (AREA)
  • Biophysics (AREA)
  • Biomedical Technology (AREA)
  • Health & Medical Sciences (AREA)
  • Computational Linguistics (AREA)
  • General Health & Medical Sciences (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • Mathematical Physics (AREA)
  • Image Analysis (AREA)
  • Information Retrieval, Db Structures And Fs Structures Therefor (AREA)

Abstract

本申请提供了一种基于神经网络的图像聚类的方法、装置、存储介质与处理器。该方法包括获取包括多个图像的图像数据库；建立图像数据库对应的哈希列表；利用哈希列表确定图像相似度模型；根据图像相似度模型构建聚类数据库，通过获取图像数据库，再对所获取的图像数据库中的图像进行处理得到图像数据库对应的哈希列表，通过哈希列表对图像数据库中的图像进行初步筛选，根据筛选出来的图像再利用哈希列表确定图像相似度模型，这样能够加速聚类的过程，从而提高了获取聚类数据库的效率，并且，通过哈希列表先进行筛选，去除一些不相关的图像，使得图像聚类效果较好，即可实现较好的图像聚类效果。

Figure 201911286423

The present application provides a method, device, storage medium and processor for image clustering based on neural network. The method includes acquiring an image database including a plurality of images; establishing a hash list corresponding to the image database; using the hash list to determine an image similarity model; constructing a clustering database according to the image similarity model; The image in the image database is processed to obtain the corresponding hash list of the image database, the images in the image database are preliminarily screened through the hash list, and the image similarity model is determined by using the hash list according to the filtered images, which can speed up The process of clustering improves the efficiency of obtaining the clustering database, and firstly filters through the hash list to remove some irrelevant images, so that the image clustering effect is better, and a better image clustering effect can be achieved. .

Figure 201911286423

Description

Image clustering method and device based on neural network and storage medium
Technical Field
The present application relates to the field of image processing, and in particular, to a method, an apparatus, a storage medium, and a processor for image clustering based on a neural network.
Background
Image processing is a popular field for artificial intelligence development, and with the popularization of mobile internet and the rise of internet of things, the traditional retrieval mode of people is more and more abundant. The current relatively mature emerging retrieval means is image retrieval. The realization of the high-efficiency clustering of large-scale image data is an important guarantee for the success of image retrieval.
The current accepted and effective image clustering process is to firstly reduce the dimension of the image, extract the low-dimensional features, and then perform clustering according to the extracted features. There are two main methods for feature extraction: feature engineering and neural networks. The feature engineering is that some interpretable image feature data, such as shooting time, image frequency, image gray level histogram, etc., are calculated and obtained through a specific formula, and has the defects that effective features are difficult to find manually, and the well-allocated features are changed as the image condition is changed, so that the flexibility is poor. The neural network transforms the picture into a feature vector through the neural network, the specific meaning of each dimension of the feature vector cannot be explained, and the neural network is diversified in selection. The clustering algorithm is commonly used at present in K-means, DB-Scan and the like, and two biggest challenges of the clustering algorithm are the operation speed and the clustering effect.
The above information disclosed in this background section is only for enhancement of understanding of the background of the technology described herein and, therefore, certain information may be included in the background that does not form the prior art that is already known in this country to a person of ordinary skill in the art.
Disclosure of Invention
The present application mainly aims to provide a method, an apparatus, a storage medium, and a processor for image clustering based on a neural network, so as to solve the problem of slow image clustering in the prior art.
In order to achieve the above object, according to one aspect of the present application, there is provided a neural network-based image clustering method, including: acquiring an image database comprising a plurality of images; establishing a hash list corresponding to the image database; determining an image similarity model by using the hash list; and constructing a clustering database according to the image similarity model.
Further, establishing a hash list corresponding to the image database, including: preprocessing each image; and establishing a hash list corresponding to the image database according to the preprocessed images.
Further, establishing a hash list corresponding to the image database, including: dividing pixel values of each pixel point of each image on a plurality of image channels to obtain a plurality of divided regions, wherein the pixel values of adjacent divided regions in one image channel are overlapped; respectively representing each divided region of each image channel by using a preset identifier, and determining the identifier of each pixel point on each image channel; determining a joint area identifier of each pixel point according to a plurality of identifiers of each pixel point on a plurality of image channels, wherein the joint area identifier is a hash value; and establishing a hash list corresponding to the image database according to the joint area identifier.
Further, after the joint area identifier of each pixel point is formed, a hash list corresponding to the image database is established, which further includes: determining the gray distribution characteristics of the image according to the joint area identification of each pixel point; and constructing the hash list according to the gray distribution characteristics of each image.
Further, determining the gray distribution characteristics of the image according to the joint region identifier of each pixel point includes: determining a frequency of occurrence of each of the joint region identifiers in each of the images; determining the likelihood probability of each joint region identifier according to the frequency and the total number of pixel points of the image; and determining the gray distribution characteristics of the image according to the likelihood probability, the ID and the joint area identification of the image.
Further, after determining the gray distribution characteristics of the image, establishing a hash list corresponding to the image database, further comprising: and establishing a hash list corresponding to the image database according to the gray distribution characteristics of the images in the image database.
Furthermore, the number of the image channels is three, namely a first channel, a second channel and a third channel.
Further, determining an image similarity model using the hash list includes: constructing an image characteristic pre-training model; and constructing the image similarity model according to the image feature pre-training model.
Further, the image similarity model includes an objective function, the objective function is
Figure BDA0002318096170000021
Wherein D ismk＝||vm-vk||2α is a hyper-parameter, vm and vk are feature vectors corresponding to the mth image and the kth image, respectively, m is 1 or 2, k is a positive integer greater than or equal to 1, and m is not equal to k, N is a positive integer greater than or equal to 3.
Further, constructing the image similarity model according to the image feature pre-training model comprises: under the condition that the image feature pre-training model is a feature project, adding a full link layer on the image feature pre-training model to determine the image similarity model; and under the condition that the image feature pre-training model is not a feature engineering, adding a preset neural network layer on the image feature pre-training model to form a preset image similarity model, training the preset image similarity model, and determining the image similarity model.
Further, constructing a clustering database according to the image similarity model, comprising: determining a feature vector of each image to be clustered according to the image similarity model; determining the images of the same category according to the feature vectors; and constructing a clustering database by adopting the images of the same category.
Further, the images of the same category are the images with the same category number, the same category ID, and the same central feature of the category, and after the clustering database is constructed by using the images of the same category, the method further includes: determining a feature vector of each image to be clustered; comparing each feature vector with the central features of each category in the clustering database to obtain a first comparison result; determining the category of each image to be clustered according to whether the first comparison result is in a first preset range; adding each image to be clustered into the corresponding category, and updating the central features of each category; comparing the updated central features of any two categories to obtain a second comparison result; fusing any two of the categories if the second comparison result is within a second predetermined range; and updating the clustering database according to the fused categories.
According to another aspect of the present application, there is provided an apparatus for neural network-based image clustering, including: an acquisition unit configured to acquire an image database including a plurality of images; the establishing unit is used for establishing a hash list corresponding to the image database; a first determining unit, configured to determine an image similarity model using the hash list; and the construction unit is used for constructing a clustering database according to the similarity model.
According to yet another aspect of the application, there is provided a storage medium comprising a stored program, wherein the program performs any one of the methods.
According to yet another aspect of the application, a processor for running a program is provided, wherein the program when running performs any of the methods.
By the technical scheme, the image clustering method and the image clustering system have the advantages that the image database is obtained, the images in the obtained image database are processed to obtain the hash lists corresponding to the image database, the images in the image database are preliminarily screened through the hash lists, the images screened out are reused, the hash lists determine the image similarity models, the clustering process can be accelerated, the efficiency of obtaining the clustering database is improved, the hash lists are screened firstly, some irrelevant images are removed, the image clustering effect is good, and the good image clustering effect can be achieved.
Drawings
The accompanying drawings, which are incorporated in and constitute a part of this application, illustrate embodiments of the application and, together with the description, serve to explain the application and are not intended to limit the application. In the drawings:
FIG. 1 illustrates a flow diagram of a method of neural network-based image clustering in accordance with an embodiment of the present application; and
fig. 2 shows a schematic diagram of an apparatus for neural network-based image clustering according to an embodiment of the present application.
Detailed Description
It should be noted that the embodiments and features of the embodiments in the present application may be combined with each other without conflict. The present application will be described in detail below with reference to the embodiments with reference to the attached drawings.
In order to make the technical solutions better understood by those skilled in the art, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application, and it is obvious that the described embodiments are only partial embodiments of the present application, but not all embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present application.
It should be noted that the terms "first," "second," and the like in the description and claims of this application and in the drawings described above are used for distinguishing between similar elements and not necessarily for describing a particular sequential or chronological order. It should be understood that the data so used may be interchanged under appropriate circumstances such that embodiments of the application described herein may be used. Furthermore, the terms "comprises," "comprising," and "having," and any variations thereof, are intended to cover a non-exclusive inclusion, such that a process, method, system, article, or apparatus that comprises a list of steps or elements is not necessarily limited to those steps or elements expressly listed, but may include other steps or elements not expressly listed or inherent to such process, method, article, or apparatus.
It will be understood that when an element such as a layer, film, region, or substrate is referred to as being "on" another element, it can be directly on the other element or intervening elements may also be present. Also, in the specification and claims, when an element is described as being "connected" to another element, the element may be "directly connected" to the other element or "connected" to the other element through a third element.
For convenience of description, some terms or expressions referred to in the embodiments of the present application are explained below:
and (3) hashing: it is a function of transforming an input of arbitrary length into a fixed length output, the output value is a hash value, the transformation is a compression mapping, that is, the space of the hash value is usually much smaller than that of the input, different inputs may be hashed into the same output, so it is impossible to determine the unique input value from the hash value, which is simply a function of compressing a message of arbitrary length into a message digest of a certain fixed length.
As introduced in the background art, the image clustering efficiency in the prior art is low, and in order to solve the problem of poor image clustering effect, embodiments of the present application provide a method, an apparatus, a storage medium, and a processor for image clustering based on a neural network.
According to an exemplary embodiment of the present application, a method of image clustering based on a neural network is provided.
Fig. 1 is a flow chart of a method of neural network-based image clustering according to an embodiment of the present application. As shown in fig. 1, the method comprises the steps of:
step S101, obtaining an image database comprising a plurality of images;
step S102, establishing a hash list corresponding to the image database;
step S103, determining an image similarity model by utilizing the hash list;
and step S104, constructing a clustering database according to the image similarity model.
In the above scheme, through obtaining the image database, handle the hash list that obtains the image database correspondence to the image in the image database who obtains again, carry out preliminary screening through the image in the hash list to the image database, recycle above-mentioned hash list according to the image of screening out and confirm the image similarity model, the process of clustering so can be accelerated, thereby the efficiency of obtaining the clustering database has been improved, and, filter earlier through the hash list, get rid of some irrelevant images, make image clustering effect better, namely, can realize better image clustering effect.
It should be noted that the steps illustrated in the flowcharts of the figures may be performed in a computer system such as a set of computer-executable instructions and that, although a logical order is illustrated in the flowcharts, in some cases, the steps illustrated or described may be performed in an order different than presented herein.
In an embodiment of the present application, the establishing of the hash list corresponding to the image database includes: preprocessing each image; and establishing a hash list corresponding to the image database according to the preprocessed images, and preprocessing the images, wherein the preprocessing can be normalization processing, and Z-score normalization and other methods can be adopted, so that all the images are unified in one range, the hash list is conveniently established, and the construction of a clustering database is facilitated.
In an embodiment of the present application, the establishing of the hash list corresponding to the image database includes: dividing pixel values of each pixel point of each image in a plurality of image channels to obtain a plurality of divided regions, wherein the pixel values of adjacent divided regions in one image channel are overlapped; respectively representing each divided region of each image channel by using a preset identifier, and determining the identifier of each pixel point on each image channel; determining a joint area identifier of each pixel point according to a plurality of identifiers of each pixel point on a plurality of image channels, wherein the joint area identifier is a hash value; and establishing a hash list corresponding to the image database according to the joint area identifier, wherein in a specific implementation mode, assuming that the image is a three-channel image, the image has 3 channels, the value of a pixel point on each channel is 0-255, and the value of 0-255 on each channel is divided according to a certain interval to obtain a hash key. For example, with a simple uniform overlap division method, as follows:
a first channel: 0 to 31,16 to 47,32 to 63,48 to 79.
A second channel: 0 to 31,16 to 47,32 to 63,48 to 79.
A third channel: 0 to 31,16 to 47,32 to 63,48 to 79.
Each channel is divided into 16 zones with an overlap between adjacent zones. The three channels have 16 × 16 ═ 4096 union regions. We use the letters a-P to denote the zones of each channel separately, and the joint zone identification of the three channels is the hash value. For example, if a pixel point is given as (10,40,8), the first channel is identified as 'a', the second channel belongs to two areas 'B', 'C', and the third channel belongs to 'a', the joint areas of the pixel point are identified as 'ABA' and 'ACA', the joint area is identified as a hash value, and other identification methods may be selected.
In an embodiment of the present application, after the joint area identifier of each pixel point is formed, the hash list corresponding to the image database is established, which further includes: determining the gray distribution characteristics of the image according to the joint area identification of each pixel point; and constructing the hash list according to the gray distribution characteristics of the images, namely determining the gray distribution characteristics of the images according to the joint area identification of each pixel point, and further constructing the hash list according to the gray distribution characteristics of the images.
In an embodiment of the application, determining the gray distribution characteristics of the image according to the joint area identifier of each pixel point includes: determining the frequency of occurrence of each joint region mark in each image; determining the likelihood probability of each joint area mark according to the frequency and the total number of pixel points of the image; determining the gray distribution characteristics of the image according to the likelihood probability, the ID and the joint area identification of the image, in a specific implementation mode, giving an image, calculating the joint area identification of each pixel point, counting the occurrence frequency of all the joint area identifications, and dividing the frequency by the total number of the pixel points of the image to obtain the likelihood probability P of the pixel point occurring in each joint area. For example, there is only one 100 x 100 image in the database, with image ID 3532, where half of the pixel values are (0,0,0) and the other half are (255 ). The gray scale distribution characteristic of this image is then:
{‘AAA’:{ID:3532,P:0.5},‘PPP’:{ID:3532,P:0.5}}
in an embodiment of the present application, after determining the gray scale distribution characteristics of the image, establishing a hash list corresponding to the image database, further includes: establishing a hash list corresponding to the image database according to the gray scale distribution characteristics of each image in the image database, constructing a hash table, summarizing gray scale distribution information of all images, and constructing a hash table, for example, the hash table H is constructed as follows,
H[‘AAA’]＝[{ID:3532,P:0.5,CID:NULL},
{ID:22,P:0.23,CID:NULL},...],
H[‘BAA’]＝[{ID:23451,P:0.045,CID:NULL},...],....
H[‘PPP’]＝[{ID＝3532,rate＝0.5,CID＝NULL},...]
in one embodiment of the present application, there are three image channels, namely, a first channel, a second channel and a third channel, and in the case where the image is a color image, there are three image channels, namely, a red (R) channel, a blue (B) channel and a green (G) channel.
An embodiment of the present application includes a hash table retrieval method, specifically, a given image x is determined, a gray distribution feature of the given image x is determined, a hash value is determined according to the gray distribution feature, and all images having the same gray area as the given image x are retrieved. For example, the gray distribution of x is characterized by { 'AAA': ID:3532, P:0.5}, 'PPP': ID:3532, P:0.5} }, and the search result is:
[{ID:3532,P:0.5,CID:NULL},
{ID:22,P:0.23,CID:NULL},...,
{ID＝3532,rate＝0.5,CID＝NULL},...]
an embodiment of the present application, determining an image similarity model by using the hash list includes: constructing an image characteristic pre-training model; and constructing the image similarity model according to the image feature pre-training model, namely constructing the image feature pre-training model firstly, and constructing the image similarity model on the basis of the image feature pre-training model so as to ensure the accuracy of the image similarity model.
In an embodiment of the present application, the image similarity model includes an objective function, and the objective function is
Figure BDA0002318096170000061
Wherein D ismk＝||vm-vk||2α denotes a hyper-parameter, vm and vk denote feature vectors corresponding to the mth image and the kth image, respectively, m is 1 or 2, k denotes a positive integer greater than or equal to 1, m ≠ k, N denotes a positive integer greater than or equal to 3, and two pictures x1 and x2 respectively input S to obtain vector representations v1 and v2, and the similarity D12 between x1 and x2 is defined as the distance between v1 and v2, and may be represented by, for example, a 2 norm:
D12＝||v1-v2||2
the hyper-parameter α is set, and if D12< α, x1 and x2 are considered similar.
In the training process, each batch needs special organization training data, namely, images of the same group are extracted in proportion randomly, other images are from different groups, and each batch is guaranteed to contain the images of the same group. For example, assuming that the number of images of a batch is 128, the first two images are designated as the same type of image, and the last 126 images are randomly extracted from the rest of other categories (assuming that the number of categories is greater than 126), the objective function is
Figure BDA0002318096170000062
Whether the extracted images are of the same class is evaluated by calculating the size of an objective function (loss function).
According to an embodiment of the present application, constructing the image similarity model according to the image feature pre-training model includes: under the condition that the image characteristic pre-training model is a characteristic project, adding a full link layer on the image characteristic pre-training model to determine the image similarity model; and under the condition that the image feature pre-training model is not a feature engineering, adding a preset neural network layer on the image feature pre-training model to form a preset image similarity model, training the preset image similarity model, and determining the image similarity model so as to realize the accurate construction of the image feature pre-training model.
In an embodiment of the present application, constructing a clustering database according to the image similarity model includes: determining the characteristic vector of each image to be clustered according to the image similarity model; determining the images of the same category according to the feature vectors; adopting the images of the same category to construct a clustering database, using the feature vectors to represent the images, further obtaining the images of the same category according to the feature vectors, further adopting the image clustering database of the same category, in a specific implementation mode, using a trained similarity model, calculating the feature vectors of each image, then averaging to obtain the central feature f of the images of a certain category, if some images are clustered into a category, then each category information comprises a category number CID, all image IDs of the category and the central feature f of the category, the clustering database is designed as follows,
{‘CID1’:{images:[ID11,ID12,...]},f:[0.11,0.04,...]},
‘CID2’:{images:[ID21,ID22,...]},f:[0.7,0.001,...]},...}
in an embodiment of the present application, after the images of the same category are the images with the same category number, the same ID, and the same central feature, and a clustering database is constructed by using the images of the same category, the method further includes determining a feature vector of each of the images to be clustered, comparing each of the feature vectors with the central feature of each of the categories in the clustering database to obtain a first comparison result, determining the category of each of the images to be clustered according to whether the first comparison result is within a first predetermined range, adding each of the images to be clustered to the corresponding category, updating the central feature of each of the categories, comparing the updated central features of any two of the categories to obtain a second comparison result, fusing any two of the categories when the second comparison result is within a second predetermined range, updating the clustering database according to the fused categories, wherein N images to be clustered in the clustering database are set, each image is represented by xi, and the total number of the clustering database is α, and the total number of the clustering database is set as α:
Figure BDA0002318096170000071
Figure BDA0002318096170000081
and clustering once or for multiple times until a more accurate clustering database is obtained, so as to realize accurate clustering of the images.
The embodiment of the present application further provides a device for image clustering based on a neural network, and it should be noted that the device for image clustering based on a neural network according to the embodiment of the present application can be used for executing the method for image clustering based on a neural network provided by the embodiment of the present application. The following describes an apparatus for image clustering based on a neural network according to an embodiment of the present application.
Fig. 2 is a schematic diagram of an apparatus for neural network-based image clustering according to an embodiment of the present application. As shown in fig. 2, the apparatus includes:
an acquisition unit 10 for acquiring an image database including a plurality of images;
an establishing unit 20, configured to establish a hash list corresponding to the image database;
a first determining unit 30, configured to determine an image similarity model using the hash list;
and the building unit 40 is used for building a clustering database according to the similarity model.
In the above scheme, obtain the image database through the acquisition element, handle the hash list that the image database corresponds that obtains the image database to the image in the image database who obtains again, carry out preliminary screening through the image in the hash list to the image database, first definite unit utilizes above-mentioned hash list to confirm the image similarity model according to the image of screening out again, the process of clustering can accelerate like this, thereby the efficiency of obtaining the cluster database has been improved, and, filter earlier through the hash list, get rid of some irrelevant images, make image clustering effect better, can realize better image clustering effect.
In an embodiment of the application, the establishing unit includes a preprocessing module and an establishing module, and the preprocessing module is configured to preprocess each of the images; the establishing module is used for establishing a hash list corresponding to the image database according to the preprocessed images, preprocessing the images, wherein the preprocessing can be normalization processing, and Z-score normalization and other methods can be adopted, so that all the images are unified in one range, the hash list can be conveniently established, and the clustering database can be favorably established.
In an embodiment of the present application, the creating unit includes a dividing module, a first determining module and a second determining module, where the dividing module is configured to divide pixel values of each pixel point of each of the images in a plurality of image channels to obtain a plurality of divided regions, and the pixel values of adjacent divided regions in one of the image channels overlap each other; the first determining module is used for respectively representing each divided area of each image channel by using a preset identifier and determining the identifier of each pixel point on each image channel; the second determining module is used for determining a joint area identifier of each pixel point according to a plurality of identifiers of each pixel point on a plurality of image channels, wherein the joint area identifier is a hash value; and establishing a hash list corresponding to the image database according to the joint area identifier, wherein in a specific implementation mode, assuming that the image is a three-channel image, the image has 3 channels, the value of a pixel point on each channel is 0-255, and the value of 0-255 on each channel is divided according to a certain interval to obtain a hash key. For example, with a simple uniform overlap division method, as follows:
a first channel: 0 to 31,16 to 47,32 to 63,48 to 79.
A second channel: 0 to 31,16 to 47,32 to 63,48 to 79.
A third channel: 0 to 31,16 to 47,32 to 63,48 to 79.
Each channel is divided into 16 zones with an overlap between adjacent zones. The three channels have 16 × 16 ═ 4096 union regions. We use the letters a-P to denote the zones of each channel separately, and the joint zone identification of the three channels is the hash value. For example, if a pixel point is given as (10,40,8), the first channel is identified as 'a', the second channel belongs to two areas 'B', 'C', and the third channel belongs to 'a', the joint areas of the pixel point are identified as 'ABA' and 'ACA', the joint area is identified as a hash value, and other identification methods may be selected.
In an embodiment of the present application, after the joint area identifier of each pixel point is formed, the establishing unit further includes a third determining module and a first establishing module, where the third determining module is configured to determine a gray distribution characteristic of the image according to the joint area identifier of each pixel point; the first construction module is used for constructing the Hash list according to the gray distribution characteristics of the images, namely determining the gray distribution characteristics of the images according to the joint area identification of each pixel point, and further constructing the Hash list according to the gray distribution characteristics of the images.
In an embodiment of the present application, the third determining module includes a first determining sub-module, a second determining sub-module, and a third determining sub-module, where the first determining sub-module is configured to determine a frequency of occurrence of each joint area identifier in each of the images; the second determining submodule is used for determining the likelihood probability of each joint area mark according to the frequency and the total number of pixel points of the image; and the third determining submodule is used for determining the gray distribution characteristics of the image according to the likelihood probability, the ID and the joint area identification of the image. For example, there is only one 100 x 100 image in the database, with image ID 3532, where half of the pixel values are (0,0,0) and the other half are (255 ). The gray scale distribution characteristic of this image is then:
{‘AAA’:{ID:3532,P:0.5},‘PPP’:{ID:3532,P:0.5}}
in an embodiment of the application, after determining the gray scale distribution characteristics of the images, the establishing unit is further configured to establish a hash list corresponding to the image database according to the gray scale distribution characteristics of each of the images in the image database, construct a hash table, summarize gray scale distribution information of all the images, and construct a hash table, for example, the hash table H is constructed as follows,
H[‘AAA’]＝[{ID:3532,P:0.5,CID:NULL},
{ID:22,P:0.23,CID:NULL},...],
H[‘BAA’]＝[{ID:23451,P:0.045,CID:NULL},...],....
H[‘PPP’]＝[{ID＝3532,rate＝0.5,CID＝NULL},...]
in one embodiment of the present application, there are three image channels, namely, a first channel, a second channel and a third channel, and in the case where the image is a color image, there are three image channels, namely, a red (R) channel, a blue (B) channel and a green (G) channel.
An embodiment of the present application includes a hash table retrieving module, which is specifically configured to give an image x, determine a gray distribution characteristic of the image x, determine a hash value according to the gray distribution characteristic, and retrieve all images having a same gray area as the image x. For example, the gray distribution of x is characterized by { 'AAA': ID:3532, P:0.5}, 'PPP': ID:3532, P:0.5} }, and the search result is:
[{ID:3532,P:0.5,CID:NULL},
{ID:22,P:0.23,CID:NULL},...,
{ID＝3532,rate＝0.5,CID＝NULL},...]
according to one embodiment of the application, the first determining unit comprises a second building module and a third building module, wherein the second building module is used for building an image feature pre-training model; the third construction module is used for constructing the image similarity model according to the image feature pre-training model, namely, firstly constructing the image feature pre-training model, and then constructing the image similarity model on the basis of the image feature pre-training model so as to ensure the accuracy of the image similarity model.
In an embodiment of the present application, the image similarity model includes an objective function, and the objective function is
Figure BDA0002318096170000101
Wherein D ismk＝||vm-vk||2α denotes a hyper-parameter, vm and vk denote feature vectors corresponding to the mth image and the kth image, respectively, m is 1 or 2, k denotes a positive integer greater than or equal to 1, m ≠ k, N denotes a positive integer greater than or equal to 3, and two pictures x1 and x2 respectively input S to obtain vector representations v1 and v2, and the similarity D12 between x1 and x2 is defined as the distance between v1 and v2, and may be represented by, for example, a 2 norm:
D12＝||v1-v2||2
the hyper-parameter α is set, and if D12< α, x1 and x2 are considered similar.
In the training process, each batch needs special organization training data, namely, images of the same group are extracted in proportion randomly, other images are from different groups, and each batch is guaranteed to contain the images of the same group. For example, assuming that the number of images of a batch is 128, the first two images are designated as the same type of image, and the last 126 images are randomly extracted from the rest of other categories (assuming that the number of categories is greater than 126), the objective function is
Figure BDA0002318096170000111
Whether the extracted images are of the same class is evaluated by calculating the size of an objective function (loss function).
In an embodiment of the present application, the third building module includes a fourth determining submodule and a fifth determining submodule, where the fourth determining submodule is configured to add a full link layer to the image feature pre-training model to determine the image similarity model when the image feature pre-training model is a feature engineering; and the fifth determining submodule is used for adding a preset neural network layer on the image feature pre-training model to determine a preset image similarity model under the condition that the image feature pre-training model is not a feature project, training the preset image similarity model and determining the image similarity model so as to realize the accurate construction of the image feature pre-training model.
In an embodiment of the present application, the building unit includes a fourth determining module, a fifth determining module, and a fourth building module, where the fourth determining module is configured to determine, according to the image similarity model, a feature vector of each of the images to be clustered; the fifth determining module is used for determining the images of the same category according to the feature vectors; the fourth building module is used for building a clustering database by using the images of the same category, using the feature vectors to represent the images, further obtaining the images of the same category according to the feature vectors, further using the image clustering database of the same category, in a specific implementation mode, using a trained similarity model, calculating the feature vectors of each image, then averaging to obtain the central feature f of the images of a certain category, if some images are clustered into a category, each category information contains a category number CID, all image IDs of the category and the central feature f of the category, the clustering database is designed as follows,
{‘CID1’:{images:[ID11,ID12,...]},f:[0.11,0.04,...]},
‘CID2’:{images:[ID21,ID22,...]},f:[0.7,0.001,...]},...}
in one embodiment of the present application, the image of the same category is the image with the same category number, the same ID and the same central feature, the apparatus further includes a second determining unit, a first comparing unit, a third determining unit, a first updating unit, a second comparing unit, a merging unit and a second updating unit, wherein after the clustering database is constructed by using the images of the same category, a feature vector of each image to be clustered is determined, the first comparing unit is configured to compare each feature vector with the central feature of each category in the clustering database to obtain a first comparison result, the third determining unit is configured to determine the category of each image to be clustered according to whether the first comparison result is within a first predetermined range, the first updating unit is configured to add each image to be clustered to the corresponding category to update the central feature of each category, the second comparing unit is configured to compare the central features of any two updated categories to obtain a second comparison result, the second comparing unit is configured to set the total number of the images to be clustered in the second predetermined range, and set a parameter of the total number of the images to be merged in the clustering database after the clustering database is set as α:
Figure BDA0002318096170000112
Figure BDA0002318096170000121
and finishing the first clustering to obtain a clustering database C, and carrying out secondary or multiple clustering on the C. The secondary clustering process is as follows:
Figure BDA0002318096170000122
and clustering once or for multiple times until a more accurate clustering database is obtained, so as to realize accurate clustering of the images.
The device for image clustering based on the neural network comprises a processor and a memory, wherein the acquiring unit, the establishing unit, the first determining unit, the constructing unit and the like are stored in the memory as program units, and the processor executes the program units stored in the memory to realize corresponding functions.
The processor comprises a kernel, and the kernel calls the corresponding program unit from the memory. One or more than one kernel can be set, and a better image clustering effect is obtained by adjusting kernel parameters.
The memory may include volatile memory in a computer readable medium, Random Access Memory (RAM) and/or nonvolatile memory such as Read Only Memory (ROM) or flash memory (flash RAM), and the memory includes at least one memory chip.
An embodiment of the present invention provides a storage medium on which a program is stored, which when executed by a processor implements the above-described method for image clustering based on a neural network.
The embodiment of the invention provides a processor, which is used for running a program, wherein the method for clustering images based on a neural network is executed when the program runs.
The embodiment of the invention provides equipment, which comprises a processor, a memory and a program which is stored on the memory and can run on the processor, wherein when the processor executes the program, at least the following steps are realized:
step S101, obtaining an image database comprising a plurality of images;
step S102, establishing a hash list corresponding to the image database;
step S103, determining an image similarity model by utilizing the hash list;
and step S104, constructing a clustering database according to the image similarity model.
The device herein may be a server, a PC, a PAD, a mobile phone, etc.
The present application further provides a computer program product adapted to perform a program of initializing at least the following method steps when executed on a data processing device:
step S101, obtaining an image database comprising a plurality of images;
step S102, establishing a hash list corresponding to the image database;
step S103, determining an image similarity model by utilizing the hash list;
and step S104, constructing a clustering database according to the image similarity model.
As will be appreciated by one skilled in the art, embodiments of the present application may be provided as a method, system, or computer program product. Accordingly, the present application may take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Furthermore, the present application may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, and the like) having computer-usable program code embodied therein.
The present application is described with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the application. It will be understood that each flow and/or block of the flow diagrams and/or block diagrams, and combinations of flows and/or blocks in the flow diagrams and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instruction means which implement the function specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
In a typical configuration, a computing device includes one or more processors (CPUs), input/output interfaces, network interfaces, and memory.
The memory may include forms of volatile memory in a computer readable medium, Random Access Memory (RAM) and/or non-volatile memory, such as Read Only Memory (ROM) or flash memory (flash RAM). The memory is an example of a computer-readable medium.
Computer-readable media, including both non-transitory and non-transitory, removable and non-removable media, may implement information storage by any method or technology. The information may be computer readable instructions, data structures, modules of a program, or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), Static Random Access Memory (SRAM), Dynamic Random Access Memory (DRAM), other types of Random Access Memory (RAM), Read Only Memory (ROM), Electrically Erasable Programmable Read Only Memory (EEPROM), flash memory or other memory technology, compact disc read only memory (CD-ROM), Digital Versatile Discs (DVD) or other optical storage, magnetic cassettes, magnetic tape magnetic disk storage or other magnetic storage devices, or any other non-transmission medium that can be used to store information that can be accessed by a computing device. As defined herein, a computer readable medium does not include a transitory computer readable medium such as a modulated data signal and a carrier wave.
It should also be noted that the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising an … …" does not exclude the presence of other identical elements in the process, method, article, or apparatus that comprises the element.
From the above description, it can be seen that the above-described embodiments of the present application achieve the following technical effects:
1) the utility model provides a method for image clustering based on neural network, through obtaining the image database, handle the image in the image database who obtains again and obtain the hash list that the image database corresponds, carry out preliminary screening through the image in the hash list to the image database, recycle above-mentioned hash list according to the image of screening out and confirm the image similarity model, the process of clustering can be quickened like this, thereby the efficiency of obtaining the cluster database has been improved, and, filter earlier through the hash list, get rid of some irrelevant images, make image clustering effect better, can realize better image clustering effect.
2) The utility model provides a device of image clustering based on neural network, through the acquisition element acquisition image database, handle the hash list that the image database corresponds that obtains the image database to the image in the image database who obtains again, carry out preliminary screening through the image in the hash list to the image database, the first unit of confirming of rethread utilizes above-mentioned hash list to confirm the image similarity model, the image similarity model here is improved image similarity model, and then the construction unit constructs the clustering database according to improved image similarity model, can realize better image clustering effect.
The above description is only a preferred embodiment of the present application and is not intended to limit the present application, and various modifications and changes may be made by those skilled in the art. Any modification, equivalent replacement, improvement and the like made within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims (15)

1. A method of image clustering based on a neural network, comprising:
acquiring an image database comprising a plurality of images;
establishing a hash list corresponding to the image database;
determining an image similarity model by using the hash list;
and constructing a clustering database according to the image similarity model.
2. The method of claim 1, wherein establishing the hash list corresponding to the image database comprises:
preprocessing each image;
and establishing a hash list corresponding to the image database according to the preprocessed images.
3. The method of claim 1, wherein establishing the hash list corresponding to the image database comprises:
dividing pixel values of each pixel point of each image on a plurality of image channels to obtain a plurality of divided regions, wherein the pixel values of adjacent divided regions in one image channel are overlapped;
respectively representing each divided region of each image channel by using a preset identifier, and determining the identifier of each pixel point on each image channel;
determining a joint area identifier of each pixel point according to a plurality of identifiers of each pixel point on a plurality of image channels, wherein the joint area identifier is a hash value;
and establishing a hash list corresponding to the image database according to the joint area identifier.
4. The method of claim 3, wherein after forming the union region identifier of each pixel point, establishing a hash list corresponding to the image database, further comprises:
determining the gray distribution characteristics of the image according to the joint area identification of each pixel point;
and constructing the hash list according to the gray distribution characteristics of each image.
5. The method of claim 4, wherein determining the gray scale distribution characteristics of the image according to the joint region identifier of each pixel point comprises:
determining a frequency of occurrence of each of the joint region identifiers in each of the images;
determining the likelihood probability of each joint region identifier according to the frequency and the total number of pixel points of the image;
and determining the gray distribution characteristics of the image according to the likelihood probability, the ID and the joint area identification of the image.
6. The method of claim 4 or 5, wherein after determining the gray scale distribution characteristics of the image, establishing a corresponding hash list of the image database further comprises:
and establishing a hash list corresponding to the image database according to the gray distribution characteristics of the images in the image database.
7. The method of any of claims 3 to 5, wherein there are three image channels, a first channel, a second channel and a third channel.
8. The method of claim 1, wherein determining an image similarity model using the hash list comprises:
constructing an image characteristic pre-training model;
and constructing the image similarity model according to the image feature pre-training model.
9. The method of claim 8, wherein the image similarity model comprises an objective function, the objective function being
Figure FDA0002318096160000021
Wherein D ismk＝||vm-vk||2α is a hyper-parameter, vm and vk are feature vectors corresponding to the mth image and the kth image, respectively, m is 1 or 2, k is a positive integer greater than or equal to 1, and m is not equal to k, N is a positive integer greater than or equal to 3.
10. The method of claim 8 or 9, wherein constructing the image similarity model from the image feature pre-training model comprises:
under the condition that the image feature pre-training model is a feature project, adding a full link layer on the image feature pre-training model to determine the image similarity model;
and under the condition that the image feature pre-training model is not a feature engineering, adding a preset neural network layer on the image feature pre-training model to form a preset image similarity model, training the preset image similarity model, and determining the image similarity model.
11. The method of claim 1, wherein constructing a clustering database from the image similarity model comprises:
determining a feature vector of each image to be clustered according to the image similarity model;
determining the images of the same category according to the feature vectors;
and constructing a clustering database by adopting the images of the same category.
12. The method according to claim 11, wherein the images of the same category are the images with the same category number, ID and central feature of the category, and after the clustering database is built by using the images of the same category, the method further comprises:
determining a feature vector of each image to be clustered;
comparing each feature vector with the central features of each category in the clustering database to obtain a first comparison result;
determining the category of each image to be clustered according to whether the first comparison result is in a first preset range;
adding each image to be clustered into the corresponding category, and updating the central features of each category;
comparing the updated central features of any two categories to obtain a second comparison result;
fusing any two of the categories if the second comparison result is within a second predetermined range;
and updating the clustering database according to the fused categories.
13. An apparatus for image clustering based on a neural network, comprising:
an acquisition unit configured to acquire an image database including a plurality of images;
the establishing unit is used for establishing a hash list corresponding to the image database;
a first determining unit, configured to determine an image similarity model using the hash list;
and the construction unit is used for constructing a clustering database according to the similarity model.
14. A storage medium, characterized in that the storage medium comprises a stored program, wherein the program performs the method of any one of claims 1 to 12.
15. A processor, characterized in that the processor is configured to run a program, wherein the program when running performs the method of any of claims 1 to 12.
CN201911286423.3A 2019-12-13 2019-12-13 Image clustering method, device and storage medium based on neural network Active CN111079828B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201911286423.3A CN111079828B (en) 2019-12-13 2019-12-13 Image clustering method, device and storage medium based on neural network

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201911286423.3A CN111079828B (en) 2019-12-13 2019-12-13 Image clustering method, device and storage medium based on neural network

Publications (2)

Publication Number Publication Date
CN111079828A true CN111079828A (en) 2020-04-28
CN111079828B CN111079828B (en) 2023-10-20

Family

ID=70314557

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201911286423.3A Active CN111079828B (en) 2019-12-13 2019-12-13 Image clustering method, device and storage medium based on neural network

Country Status (1)

Country Link
CN (1) CN111079828B (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN118691864A (en) * 2024-08-23 2024-09-24 浙江大华技术股份有限公司 A target clustering method, device, equipment and medium

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP2011221689A (en) * 2010-04-07 2011-11-04 Sony Corp Learning device, learning method, recognition device, recognition method, and program
CN106815362A (en) * 2017-01-22 2017-06-09 福州大学 One kind is based on KPCA multilist thumbnail Hash search methods
US20190005657A1 (en) * 2017-06-30 2019-01-03 Baidu Online Network Technology (Beijing) Co., Ltd . Multiple targets-tracking method and apparatus, device and storage medium
CN109918532A (en) * 2019-03-08 2019-06-21 苏州大学 Image retrieval method, apparatus, device, and computer-readable storage medium

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP2011221689A (en) * 2010-04-07 2011-11-04 Sony Corp Learning device, learning method, recognition device, recognition method, and program
CN106815362A (en) * 2017-01-22 2017-06-09 福州大学 One kind is based on KPCA multilist thumbnail Hash search methods
US20190005657A1 (en) * 2017-06-30 2019-01-03 Baidu Online Network Technology (Beijing) Co., Ltd . Multiple targets-tracking method and apparatus, device and storage medium
CN109918532A (en) * 2019-03-08 2019-06-21 苏州大学 Image retrieval method, apparatus, device, and computer-readable storage medium

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
刘冶;潘炎;夏榕楷;刘荻;印鉴;: "FP-CNNH:一种基于深度卷积神经网络的快速图像哈希算法" *
徐名刘;徐昆;: "一种基于相对排序的有监督哈希算法" *

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN118691864A (en) * 2024-08-23 2024-09-24 浙江大华技术股份有限公司 A target clustering method, device, equipment and medium

Also Published As

Publication number Publication date
CN111079828B (en) 2023-10-20

Similar Documents

Publication Publication Date Title
Waldamichael et al. Coffee disease detection using a robust HSV color‐based segmentation and transfer learning for use on smartphones
US20130195361A1 (en) Image index generation based on similarities of image features
WO2017181892A1 (en) Foreground segmentation method and device
CN114333062B (en) Person Re-identification Model Training Method Based on Heterogeneous Dual Network and Feature Consistency
CN114023397B (en) Drug redirection model generation method and device, storage medium and computer equipment
US11074274B2 (en) Large scale social graph segmentation
CN113297420A (en) Video image processing method and device, storage medium and electronic equipment
CN111741331B (en) Video clip processing method, device, storage medium and equipment
CN114330584A (en) Data clustering method and device, storage medium and electronic equipment
CN109754023A (en) A Novel Decision Tree Classification Method Based on J Divergence
CN114449342A (en) Video recommendation method and device, computer readable storage medium and computer equipment
CN117312549A (en) Work order processing methods, devices and electronic equipment
CN112749288B (en) Multimedia data processing method, device, server and storage medium
CN111079828A (en) Method, device and storage medium for image clustering based on neural network
CN111309946B (en) Established file optimization method and device
US12242803B2 (en) Hierarchical ontology matching with self-supervision
CN114495175B (en) Pedestrian re-identification model creation method, device, equipment and storage medium
CN113377929B (en) Unsupervised clustering method, unsupervised clustering device and unsupervised clustering system for special terms
CN118035800A (en) Model training method, device, equipment and storage medium
CN112614134B (en) Image segmentation method, device, electronic device and storage medium
Xia et al. Face clustering in photo album
WO2025113058A1 (en) Model training method, apparatus, and device based on scene clustering, and storage medium
CN110839242A (en) Abnormal number identification method and device
CN107122795A (en) A kind of pedestrian integrated based on coring feature and stochastic subspace discrimination method again
CN107193979A (en) A kind of method of homologous picture retrieval

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
TA01 Transfer of patent application right
TA01 Transfer of patent application right

Effective date of registration: 20200803

Address after: 1608, 14 / F, No. 65, Beisihuan West Road, Haidian District, Beijing 100080

Applicant after: BEIJING INTERNETWARE Ltd.

Address before: No. 603, floor 6, No. 9, Shangdi 9th Street, Haidian District, Beijing 100085

Applicant before: Smart Shenzhou (Beijing) Technology Co.,Ltd.

GR01 Patent grant
GR01 Patent grant
CP03 Change of name, title or address
CP03 Change of name, title or address

Address after: 100080 1608, 14th floor, No. 65, North Fourth Ring Road West, Haidian District, Beijing

Patentee after: Shenqi Digital Co.,Ltd.

Country or region after: China

Address before: 100080 1608, 14th floor, No. 65, North Fourth Ring Road West, Haidian District, Beijing

Patentee before: BEIJING INTERNETWARE Ltd.

Country or region before: China