CN108170849B - Image classification marking method for partition diffusion - Google Patents

Abstract

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

Description

Image classification marking method for partition diffusion

Technical Field

The invention belongs to the field of image recognition, and particularly relates to a picture classification and marking method based on partition diffusion.

Background

With the development of society, the interaction between people is more and more, and people can chat, send information, send pictures and the like through mobile phones. Meanwhile, as the functions of the mobile phone are gradually enhanced, a user can obtain various photos and pictures through mobile phone photographing, various social applications and web browsing, so that a large number of pictures of different types and contents can be stored in the mobile phone of the user.

When a user browses pictures in a mobile phone, the user can only browse the pictures according to the shooting sequence or the picture acquisition date basically, and the requirement of the user for conveniently browsing specific pictures cannot be met. Meanwhile, when the pictures need to be classified and sorted, the user cannot conveniently finish the classification and sorting operation of the pictures on the mobile phone, and only after the pictures are transmitted to the computer in batches, the classification and sorting operation is carried out; this results in inefficient image classification and browsing. The recognition algorithms adopted by the existing image classification method are approximately the same but have the same problem that the image cannot be effectively and quickly recognized when the image is large, and the hardware consumption is high and the processing speed is low when the image is too large and the pixel is too large. Therefore, the application provides a new technical scheme for effectively dividing the picture and simultaneously identifying and classifying the labels in a multithreading manner so as to solve the problems in the prior art.

Disclosure of Invention

The present invention is directed to solving at least one of the problems of the prior art. Therefore, the invention provides a picture classification marking method for partition diffusion, which comprises the following steps:

and automatically configuring corresponding classification labels for the pre-stored pictures respectively, wherein the configured classification labels are selected from at least one preset classification label.

Further, the step of automatically configuring the corresponding classification label for the pre-stored picture specifically includes uniformly dividing the picture into AxB sub-regions, and then performing image recognition on each sub-region to obtain the classification label.

Further, the automatically configuring the corresponding classification label further includes:

combining and identifying images of adjacent subareas of the divided AxB subareas to obtain the classification label;

wherein the combined identification comprises selecting one of the partitions, x_i,jWhere i e (1, A), then x from the partition_i,jStarting from, the partition x is expanded in four directions_i,jAnd simultaneously detecting the chroma and/or the gray scale of the extended adjacent subarea, and stopping continuously extending to the peripheral subareas until the chroma or the gray scale exceeds a threshold value.

Further, the threshold is the partition x_i,jAverage value of chromaticity or gray level on the boundary line with the adjacent partition.

Drawings

The invention will be further understood from the following description 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 drawings, like reference numerals designate corresponding parts throughout the different views.

FIG. 1 is a flow diagram of class tagging according to an embodiment of the present invention.

Detailed Description

In order to make the objects, technical solutions and advantages of the present invention more clearly apparent, the present invention is further described in detail below with reference to the accompanying drawings and embodiments thereof; it should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention. Other systems, methods, and/or features of the present embodiments will become apparent to those skilled in the art upon review of 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 accompanying claims. Additional features of the disclosed embodiments are described in, and will be apparent from, the detailed description that follows.

The first embodiment.

The embodiment provides a picture classification marking method for partition diffusion, which comprises the following steps:

and automatically configuring corresponding classification labels for the pre-stored pictures respectively, wherein the configured classification labels are selected from at least one preset classification label.

Further, the step of automatically configuring the corresponding classification label for the pre-stored picture specifically includes uniformly dividing the picture into AxB sub-regions, and then performing image recognition on each sub-region to obtain the classification label.

Further, the automatically configuring the corresponding classification label further includes:

combining and identifying images of adjacent subareas of the divided AxB subareas to obtain the classification label;

wherein the combined identification comprises selecting one of the partitions, x_i,jWhere i e (1, A), then x from the partition_i,jStarting from, the partition x is expanded in four directions_i,jAnd simultaneously detecting the chroma and/or the gray scale of the extended adjacent subarea, and stopping continuously extending to the peripheral subareas until the chroma or the gray scale exceeds a threshold value.

Further, the threshold is the scoreZone x_i,jAverage value of chromaticity or gray level on the boundary line with the adjacent partition.

The specific algorithm adopted for the picture identification or the combined identification is as follows:

foreach image X_ndo

sampling the theme distribution of labels theta-Dirichlet (alpha); theta is a K-dimensional Dirichlet distribution parameterized by alpha

for each label Y_ni of image X_ndo

Assigning z to a topic_iSampling by Multinominal (θ)

From subject z_iIs/are as follows

Get a label

Is X_nCalculating a label prior:

wherein

Is Y_nMiddle y_i

The number of the particles; xi in the training process_ni＝0,η>0; xi in the test procedure_ni>0,η>0。

For labels theta 'to Dirichlet (. | alpha'_n) Sampling the topic distribution; theta 'is alpha'_nParameterized L-dimensional Dirichlet distribution

for each instance x_ni ofX_ndo

Assigning v to a label_iSampling by Multinominal (θ')

From

Taking an example;

label v_iIs a C-dimensional polynomial

for each tag t_ni in T_n of image X_ndo

Assign g to the label_iSampling by Multinominal (θ')

Slave tag g_iIs/are as follows

Get a mark

In the above algorithm, y ═ y₁,y₂,…,y_LY denotes a set of L labels, T ═ T₁,t₂,…,t_TRepresents a set of T user identities. With D { ([ X) { (]₁,T₁],Y₁),…，([X_N,T_N],Y_N) Denotes a training set of N samples, where X_n＝{x_n1,x_n2,…,x_nMnIs a group having M_nA packet of instances, T_n＝{t_n1,t_n2,…,t_nGnIs a radical having G_nSet of individual user identities, and Y_n＝{y_n1,y_n2,…,y_nLnIs L from the Y set_nA set of individual tags. The learning machine, which generates an image (or image region) based instance X and user id T (if any) for labeling, performs clustering in visual feature space to establish a prototype set C ═ C₁,c₂,…,c_C}. Wherein x_iIs a vector of size C, where x_i,cIs that prototype c occurs at x_iThe number of times (1). It should be noted that the above recognition algorithm is only a successful example, and in the practical application of the method, other recognition methods can be replaced by those skilled in the art, and the main inventive point or innovative point of the present invention or the present embodiment lies in the recognition method of partition diffusion, i.e. the architecture of multi-thread synchronous recognition.

Example two.

The embodiment provides a picture classification marking method for partition diffusion, which comprises the following steps:

and automatically configuring corresponding classification labels for the pre-stored pictures respectively, wherein the configured classification labels are selected from at least one preset classification label.

Further, the step of automatically configuring the corresponding classification label for the pre-stored picture specifically includes dividing the picture into AxB partitions uniformly, and then performing image recognition on each partition to obtain the classification label, where the number of the partitions is proportionally distributed according to the aspect ratio of the picture, and the specific number is set according to the hardware configuration of the hardware device implementing the method.

Further, the automatically configuring the corresponding classification label further includes:

combining and identifying images of adjacent subareas of the divided AxB subareas to obtain the classification label;

wherein the combined identification comprises selecting one of the partitions, x_i,jWhere i e (1, A), then x from the partition_i,jStarting from, expanding the partition x_i,jAnd simultaneously detecting the chromaticity and/or the gray scale of the adjacent partitions, and stopping continuously expanding towards the peripheral partitions until the chromaticity or the gray scale exceeds a threshold value. The initial partition can be selected by selecting the most middle partition to set or simultaneously select four or more partitions uniformly distributed on the picture, so that the partition identification process is increased by synchronously carrying out combined distribution, and the processing speed of the method is improved.

Further, the threshold is the partition x_i,jThe average value of the chromaticity or the gray scale on the boundary line with the adjacent partition, or the threshold point may be selected in the manner of a "key point" in a general picture processing method.

The specific algorithm adopted for the picture identification or the combined identification is as follows:

foreach image X_ndo

sampling the theme distribution of labels theta-Dirichlet (alpha); theta is a K-dimensional Dirichlet distribution parameterized by alpha

for each label Y_ni of image X_ndo

Assigning z to a topic_iSampling by Multinominal (θ)

From subject z_iIs/are as follows

Get a label

Is X_nCalculating a label prior:

wherein

Is Y_nMiddle y_i

The number of the particles; xi in the training process_ni＝0,η>0; xi in the test procedure_ni>0,η>0。

For labels theta 'to Dirichlet (. | alpha'_n) Sampling the topic distribution; theta 'is alpha'_nParameterized L-dimensional Dirichlet distribution

for each instance x_ni ofX_ndo

Assigning v to a label_iSampling by Multinominal (θ')

From

Taking an example;

label v_iIs a C-dimensional polynomial

for each tag t_ni in T_n of image X_ndo

Assign g to the label_iSampling by Multinominal (θ')

Slave tag g_iIs/are as follows

Get a mark

Example three.

The picture classification marking method based on the partitioned diffusion includes the steps of firstly, respectively traversing each configured classification label, performing image recognition on a target picture under the currently traversed classification label, obtaining marker features under the currently traversed classification label according to an image recognition result, and when a certain marker feature is contained in a plurality of target pictures, weighting or screening the marker features contained in each of the plurality of target pictures to obtain the marker features under the currently traversed classification label, wherein the classification label and the marker features are not in image association, and the marker features include key point positions of markers and gray values at the key point positions;

carrying out marker feature identification on the acquired picture; respectively calculating the distance values of the key point position and the gray value at the key point position in the marker feature contained in the obtained picture and the key point position and the gray value at the key point position of the marker feature under each set classification label, determining the similarity between the marker feature contained in the picture and the marker feature under each set classification label according to the distance values, and configuring the classification label, the similarity of which with the marker feature contained in the obtained picture meets the set threshold condition, to the obtained picture so as to finish the classification of the obtained picture; and storing all the acquired pictures under the same classification label in the same folder, and marking the classification label in the thumbnail of the acquired picture.

Example four.

The embodiment provides a method for classifying and marking pictures, which comprises the following steps:

and automatically configuring corresponding classification labels for the pre-stored pictures respectively, wherein the configured classification labels are selected from at least one preset classification label.

Further, the step of automatically configuring the corresponding classification label for the pre-stored picture specifically includes dividing the picture into AxB partitions uniformly, and then performing image recognition on each partition to obtain the classification label, where the number of the partitions is proportionally distributed according to the aspect ratio of the picture, and the specific number is set according to the hardware configuration of the hardware device implementing the method.

Further, the automatically configuring the corresponding classification label further includes:

combining and identifying images of adjacent subareas of the divided AxB subareas to obtain the classification label;

wherein the combined identification comprises selecting one of the partitions, x_i,jWhere i e (1, A), then x from the partition_i,jStarting from, expanding the partition x_i,jAnd detecting the chromaticity and/or the gray scale of the adjacent expanded subareas at the same time, and stopping continuously expanding towards the peripheral subareas until the chromaticity or the gray scale exceeds a threshold value, wherein the detection can detect the integral chromaticity and/or the gray scale of all the expanded subareas.

The expanding method may use a method of rotating the expanding counterclockwise or expanding the partition in eight directions at the same time. The initial partition can be selected by selecting the most middle partition to set or simultaneously select four or more partitions uniformly distributed on the picture, so that the partition identification process is increased by synchronously carrying out combined distribution, and the processing speed of the method is improved.

Further, the threshold is the partition x_i,jThe average value of the chromaticity or the gray scale on the boundary line with the adjacent partition, or the threshold point may be selected in the manner of a "key point" in a general picture processing method.

Although the invention has been described above with reference to various embodiments, it should be understood that many changes and modifications may 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 limiting, and that it be understood that it is the following claims, including all equivalents, that are intended to define the spirit and scope of this invention. The above examples are to be construed as merely illustrative and not limitative of the remainder of the disclosure. After reading the description of the invention, the skilled person can make various changes or modifications to the invention, and these equivalent changes and modifications also fall into the scope of the invention defined by the claims.

Claims (1)

1. A picture classification marking method for partition diffusion comprises the following steps:

automatically configuring corresponding classification labels for the pre-stored pictures respectively, wherein the configured classification labels are selected from at least one preset classification label; the step of automatically configuring the corresponding classification label for the pre-stored picture specifically comprises the steps of uniformly dividing the picture into AxB subareas, and then carrying out image recognition on each subarea to obtain the classification label;

and, said automatically configuring the respective classification label further comprises:

combining and identifying images of adjacent subareas of the divided AxB subareas to obtain the classification label;

wherein the combined identification comprises selecting one of the partitions, x_i,jWhere i e (1, A), j e (1, A), then x from the partition_i,jStarting from, extending the partition x in multiple directions_i,jSimultaneously detecting the chromaticity and the gray scale of the adjacent expanded partitions, and stopping continuously expanding the adjacent partitions until the chromaticity or the gray scale exceeds a threshold value;

wherein the threshold is the partition x_i,jAn average value of chromaticity or gray scale on a boundary line with an adjacent partition;

it is characterized in that the preparation method is characterized in that,

the picture identification or the combined identification specifically comprises the following steps:

sampling the picture according to theme distribution of labels theta-Dirichlet (alpha), wherein theta is K-dimensional Dirichlet distribution parameterized by alpha;

then assign z by topic_iMultinominal (θ) sampling:

from subject z_iIs/are as follows

Taking a label;

is X_nCalculating a label prior:

wherein

Is Y_nThe number of middle elements;

xi in the training process_ni＝0,η>0; xi in the test procedure_ni>0,η>0；

For labels theta 'to Dirichlet (. | alpha'_n) Sampling the topic distribution; theta 'is alpha'_nA parameterized L-dimensional Dirichlet distribution;

assigning v to a label_iMultinomial (θ') samples:

from

Taking an example;

is a label v_iA C-dimensional polynomial of (1);

assign g to the label_iMultinomial (θ') samples:

slave tag g_iIs/are as follows

Taking a mark;

wherein Y is { Y ═ Y₁,y₂,…,y_LY denotes a set of L labels, T ═ T₁,t₂,…,t_TRepresents a set of T user identities; wherein D { ([ X)₁,T₁],Y₁),…，([X_N,T_N],Y_N) Means a sample of NA training set wherein

Is a one having M_nThe packet of the one instance is,

is a one having G_nA set of individual user identities, and

is L from the set of Y_nA set of individual tags;

then, a prototype set C ═ C is established₁,c₂,…,c_C}。

Images