CN108108494A - A kind of picture classification intelligent terminal - Google Patents

Abstract

The invention discloses a method for classifying and marking pictures with partition diffusion, which includes automatically configuring corresponding classification labels for pre-saved pictures, wherein the configured classification labels are selected from at least one preset classification label. And it also provides an improved embodiment, that is, the process of classifying and marking with an accelerator by cutting and combining pictures.

Description

An intelligent terminal for image classification

technical field

The invention belongs to the field of image recognition, in particular, the invention relates to an intelligent terminal for picture classification.

Background technique

With the development of society, there are more and more interactions between people. People can chat, send messages, and send pictures through mobile phones. At the same time, due to the increasingly powerful functions of mobile phones, users will obtain various photos and pictures through mobile phone photography, various social applications and web browsing, which will cause a large number of pictures of different types and contents to be stored in the user's mobile phone. When the user browses the pictures in the mobile phone, he can basically only browse according to the shooting order or the date when the pictures were obtained, which cannot meet the needs of the user to browse specific pictures conveniently. At the same time, when the pictures need to be sorted and sorted, the user cannot conveniently complete the sorting and sorting operation of the pictures on the mobile phone, and can only transfer the pictures to the computer in batches, and then perform the sorting and sorting operation; this leads to the efficiency of picture classification and browsing are relatively low. Now the existing image classification methods adopt roughly the same recognition algorithm, but they all have the same problem, that is, they cannot effectively identify the image quickly when the image is large, and when the pixel size is too large, the image is too large. The consumption of hardware is huge and the processing speed is very low. Therefore, this application proposes a new technical solution for effectively segmenting pictures and performing multi-threaded simultaneous identification for classification and labeling to solve the existing problems in the prior art, that is, to provide an intelligent solution that can effectively solve the above technical problems. terminal.

Contents of the invention

The present invention aims to solve at least one of the technical problems existing in the prior art. To this end, the present invention provides an intelligent terminal for picture classification, the intelligent terminal is a mobile terminal, which includes: a setting unit, configured to set at least one classification label for the picture according to an instruction input by the user; a memory, It is used to pre-save the pictures acquired by the smart terminal and the classification label information; it is characterized in that the smart terminal is also equipped with a classification label determination unit for traversing the pictures in the memory, and for each picture Image recognition is performed according to each classification label configured by the sub-setting unit, and corresponding classification labels are automatically configured for the picture according to the image recognition result.

Further, the automatic configuration of corresponding classification labels for pictures specifically includes, evenly dividing the picture into AxB partitions, and then performing image recognition on each partition to obtain the classification labels.

Further, the automatic configuration of corresponding classification labels also includes:

Carrying out combined recognition on the divided AxB partitions of images of adjacent partitions to obtain the classification label;

Wherein the combined recognition includes selecting one of the partitions, i.e. x _i,j , where i∈(1,A), i∈(1,A), and then starting from the partition x _i,j , to multiple directions The partition xi _,j is extended while detecting the chromaticity and/or grayscale of the adjacent partitions to which it is extended, until the chromaticity or grayscale exceeds a threshold, then stop extending to the surrounding partitions.

Further, the threshold is an average value of chromaticity or grayscale on the boundary line between the partition xi _,j and adjacent partitions.

Description of drawings

The present invention can be further understood from the following description taken in conjunction with the accompanying drawings. The components in the figures are not necessarily to scale, emphasis instead being placed upon illustrating the principles of the embodiments. In the figures, like reference numerals designate corresponding parts in the different views.

Fig. 1 is a flow chart of classification marking in one embodiment of the present invention.

Detailed ways

In order to make the objectives, technical solutions and advantages of the present invention clearer, the present invention will be described in further detail below in conjunction with the accompanying drawings and their embodiments; it should be understood that the specific embodiments described here are only used to explain the present invention and are not intended to to limit the present invention. Other systems, methods and/or features of this embodiment will become apparent to those skilled in the art after reviewing the following detailed description. It is intended that all such additional systems, methods, features and advantages be included within this description, be within the scope of the invention, and be protected by the following claims. Additional features of the disclosed embodiments are described in, and will be apparent from, the following detailed description.

Embodiment one.

This embodiment provides an intelligent terminal for picture classification. The intelligent terminal is a mobile terminal, such as a mobile phone, a tablet computer, a camera, a video camera, a notebook computer, etc., which can obtain pictures or photos, and it includes: a setting unit for According to the input instruction, set at least one classification label for the picture; the memory is used to pre-save the picture obtained by the smart terminal and the classification label information; it is characterized in that the smart terminal also has A classification label determination unit, configured to traverse the pictures in the memory, perform image recognition on each picture according to the classification labels configured by the sub-setting unit, and automatically configure corresponding classification labels for the pictures according to the image recognition results .

Further, the automatic configuration of corresponding classification labels for pictures specifically includes, evenly dividing the picture into AxB partitions, and then performing image recognition on each partition to obtain the classification labels.

Further, the automatic configuration of corresponding classification labels also includes:

Carrying out combined recognition on the divided AxB partitions of images of adjacent partitions to obtain the classification label;

Wherein the combined recognition includes selecting one of the partitions, i.e. x _i,j , where i∈(1,A), i∈(1,A), and then starting from the partition x _i,j , to multiple directions The partition xi _,j is extended while detecting the chromaticity and/or grayscale of the adjacent partitions to which it is extended, until the chromaticity or grayscale exceeds a threshold, then stop extending to the surrounding partitions.

The threshold is an average value of chromaticity or grayscale on the boundary line between the partition xi _,j and adjacent partitions.

The specific algorithm used for image recognition or combination recognition is as follows:

foreach imageX _n do

Sampling the topic distribution of labels θ ~ Dirichlet(α); θ is a K-dimensional Dirichlet distribution parameterized by α

foreach labelY _ni ofimageX _n do

Sample the topic assignment z _i ~Multinominal(θ)

from from theme zi Take a label from

Compute label priors for X _n : in is yi in Y _n

Quantity; ξ _ni =0, η>0 during training; ξ _ni >0, η>0 during testing.

Sampling the topic distribution of labels θ′～Dirichlet(.|α′ _n ); θ′ is an L-dimensional Dirichlet distribution parameterized by α′ _n

for each instance _{x ni} of X _n do

Sample the label assignment v _i ∼ Multinominal(θ′)

from Take an instance of C-dimensional polynomial of label v _i

for each tag t _ni in T _n of image X _n do

Sampling the label assignment g _i ~Multinominal(θ')

from tag gi Take a mark from

In the above algorithm, y={y ₁ ,y ₂ ,...,y _L }, Y represents a set with L labels, represented by T={t ₁ ,t ₂ ,...,t _T } There is a collection of T user ids. Use D={([X ₁ ,T ₁ ],Y ₁ ),...,([X _N ,T _N ],Y _N )} to represent a training set with N samples, where is a bag with M _n instances, is a set of G _n user identities, and is the set of L _n labels from the Y set. The above can generate a learning machine based on the instance X of the image (or image area) and the user identifier T (if any) to perform clustering in the visual feature space to establish a prototype set C={c ₁ ,c ₂ ,...,c _C }. where _xi is a vector of size C, where xi _,c is the number of times prototype c appears in _xi . Of course, it should be noted that the above identification algorithm is only a successful example, and in the actual application of the method, other identification methods can be replaced by those skilled in the art, and the main invention points of the present invention or this embodiment or The innovation lies in the identification method of partition diffusion, that is, the architecture of multi-threaded synchronous identification.

Embodiment two.

This embodiment provides an intelligent terminal for picture classification, the intelligent terminal is configured to obtain or store pictures in advance, and it can also be used for:

Corresponding classification labels are automatically configured for the pre-saved pictures, wherein the configured classification labels are selected from at least one preset classification label.

Further, the automatic configuration of corresponding classification labels for the pre-saved pictures specifically includes, evenly dividing the picture into AxB partitions, and then performing image recognition on each partition to obtain the classification labels, the number of partitions Proportions are distributed according to the aspect ratio of the pictures, and the specific number is set according to the hardware configuration of the hardware device implementing the method.

Further, the automatic configuration of corresponding classification labels also includes:

Carrying out combined recognition on the divided AxB partitions of images of adjacent partitions to obtain the classification label;

Wherein the combination identification includes selecting one of the partitions i.e. x _i,j , where i∈(1,A), i∈(1,A), and then starting from the partition x _i,j , expanding the partition x _{i, j} simultaneously detects the chromaticity and/or grayscale of the adjacent partitions that are extended to, and stops extending to the surrounding partitions until the chromaticity or grayscale exceeds a threshold value. The method of extension can use counterclockwise rotation The method of expansion or expand the partition in eight directions at the same time. The selection of the initial partition can be set by selecting the middle partition or at the same time from four or more partitions that are evenly distributed on the picture, so that the combined allocation can be performed synchronously to make the partition recognition process more and improve the processing of this method. speed.

Further, the threshold is the average value of chromaticity or grayscale on the boundary line between the partition x _i,j and adjacent partitions, or it can be selected in the way of "key point" in the general image processing method threshold point.

The specific algorithm used for image recognition or combination recognition is as follows:

foreach image X _n do

Sampling the topic distribution of labels θ ~ Dirichlet(α); θ is a K-dimensional Dirichlet distribution parameterized by α

for each label Y _ni of image X _n do

Sample the topic assignment z _i ~Multinominal(θ)

From the topic z _i Take a label from

Compute label priors for X _n : in is y _i in Y _n

Quantity; ξ _ni =0, η>0 during training; ξ _ni >0, η>0 during testing.

Sampling the topic distribution of labels θ'～Dirichlet(.|α' _n ); θ' is an L-dimensional Dirichlet distribution parameterized by α' _n

for each instance x _ni of X _n do

Sample the label assignment v _i ∼ Multinominal(θ′)

from Take an instance of C-dimensional polynomial of label v _i

for each tag t _ni in T _n of image X _n do

Sample the label assignment gi ~ Multinominal(θ′)

from tag gi Take a mark from

Embodiment three.

This embodiment provides an intelligent terminal for classifying pictures. The intelligent terminal is configured to obtain or store pictures, and then classify and label the pictures. Image recognition is performed on the target pictures under the label, and the identifier features under the classification labels currently traversed are obtained according to the image recognition results. When a certain marker feature is included in multiple target pictures, the The features of the markers contained in each of the pictures are weighted or screened to obtain the features of the markers under the currently traversed classification label, wherein the classification label and the marker feature have no image association, and the marker The feature of the object includes the position of the key point of the marker and the gray value at the position of the key point;

Perform marker feature recognition on the acquired picture; respectively calculate the key point position in the marker feature contained in the acquired picture and the gray value at the key point position and the key of the marker feature under each set classification label The distance value of the point position and the gray value at the key point position, according to the distance value to determine the similarity between the marker features contained in the picture and the marker features under each set classification label, will be compared with the acquired The similarity of the marker features contained in the picture meets the classification label of the set threshold condition, which is allocated to the acquired picture to complete the classification of the acquired picture; all the acquired pictures under the same classification label are saved in the same folder, and mark the classification labels in the thumbnails of the acquired pictures at the same time.

Embodiment four.

This embodiment provides a picture classification label terminal, which can be used to acquire and store pictures, which includes:

Corresponding classification labels are automatically configured for the pre-saved pictures, wherein the configured classification labels are selected from at least one preset classification label.

Further, the automatic configuration of corresponding classification labels for the pre-saved pictures specifically includes, evenly dividing the picture into AxB partitions, and then performing image recognition on each partition to obtain the classification labels, the number of partitions Proportions are distributed according to the aspect ratio of the pictures, and the specific number is set according to the hardware configuration of the hardware device implementing the method.

Further, the automatic configuration of corresponding classification labels also includes:

Carrying out combined recognition on the divided AxB partitions of images of adjacent partitions to obtain the classification label;

Wherein the combination identification includes selecting one of the partitions i.e. x _i,j , where i∈(1,A), i∈(1,A), and then starting from the partition x _i,j , expanding the partition x _{i, j} simultaneously detects the chromaticity and/or grayscale of the adjacent partitions that are extended to, until the chromaticity or grayscale exceeds a threshold, it stops and continues to expand to the surrounding partitions, and the detection here can detect all extended to The overall chromaticity and/or grayscale of the partition, in this embodiment, by setting a plurality of detection points in the extension direction of the partition, for example, when extending from one partition to another partition, uniform detection along the direction of extension The chromaticity and/or grayscale values of multiple points are used as detection values.

The expansion method can use the counterclockwise rotation expansion method or expand the partition in eight directions at the same time. The selection of the initial partition can be set by selecting the middle partition or at the same time from four or more partitions that are evenly distributed on the picture, so that the combined allocation can be performed synchronously to make the partition recognition process more and improve the processing of this method. speed.

Further, the threshold is the average value of chromaticity or grayscale on the boundary line between the partition x _i,j and adjacent partitions, or it can be selected in the way of "key point" in the general image processing method threshold point.

While the invention has been described above with reference to various embodiments, it should be understood that many changes and modifications can be made without departing from the scope of the invention. It is therefore intended that the foregoing detailed description be regarded as illustrative rather than restrictive, and that it be understood that the following claims, including all equivalents, are intended to define the spirit and scope of the invention. The above embodiments should be understood as only for illustrating the present invention but not for limiting the protection scope of the present invention. After reading the content of the present invention, the skilled person can make various changes or modifications to the present invention, and these equivalent changes and modifications also fall within the scope defined by the claims of the present invention.

Claims (4)

1. A picture classification intelligent terminal, the intelligent terminal is a mobile terminal, which includes: a setting unit, configured to set at least one classification label for the pictures according to an instruction input by a user; a memory, used to save all The picture obtained by the intelligent terminal and the classification label information; it is characterized in that the intelligent terminal also has a classification label determination unit for traversing the pictures in the memory, and for each picture according to the sub-set Each classification label configured by the unit performs image recognition, and automatically configures a corresponding classification label for the picture according to the image recognition result. 2. The intelligent terminal according to claim 1, wherein the classification label determining unit automatically configures corresponding classification labels for pictures specifically comprising, dividing the picture evenly into AxB partitions, and then for each partition Image recognition is performed to obtain the classification labels. 3. The intelligent terminal according to claim 2, wherein the automatic configuration of corresponding classification labels further comprises: Carrying out combined recognition on the divided AxB partitions of images of adjacent partitions to obtain the classification label; Wherein the combined recognition includes selecting one of the partitions, i.e. x _i,j , where i∈(1,A), i∈(1,A), and then starting from the partition x _i,j , to multiple directions The partition xi _,j is extended while detecting the chromaticity and/or grayscale of the adjacent partitions to which it is extended, until the chromaticity or grayscale exceeds a threshold, then stop extending to the surrounding partitions. 4 . The intelligent terminal according to claim 3 , wherein the threshold is an average value of chromaticity or grayscale on the boundary line between the partition xi _,j and adjacent partitions.