CN111291784A - Clothing attribute identification method based on migration significance prior information - Google Patents
Clothing attribute identification method based on migration significance prior information Download PDFInfo
- Publication number
- CN111291784A CN111291784A CN202010043730.5A CN202010043730A CN111291784A CN 111291784 A CN111291784 A CN 111291784A CN 202010043730 A CN202010043730 A CN 202010043730A CN 111291784 A CN111291784 A CN 111291784A
- Authority
- CN
- China
- Prior art keywords
- clothing
- image
- significance
- prior information
- attribute
- 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.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 27
- 230000005012 migration Effects 0.000 title claims abstract description 15
- 238000013508 migration Methods 0.000 title claims abstract description 15
- 238000007781 pre-processing Methods 0.000 claims abstract description 13
- 238000013528 artificial neural network Methods 0.000 claims abstract description 10
- 238000013527 convolutional neural network Methods 0.000 claims abstract description 9
- 238000001514 detection method Methods 0.000 claims abstract description 8
- 238000012549 training Methods 0.000 claims abstract description 6
- 238000010606 normalization Methods 0.000 claims description 9
- 230000009466 transformation Effects 0.000 claims description 3
- 230000006870 function Effects 0.000 description 4
- 238000010586 diagram Methods 0.000 description 3
- 230000002708 enhancing effect Effects 0.000 description 3
- 238000011176 pooling Methods 0.000 description 3
- 230000004913 activation Effects 0.000 description 2
- 238000006243 chemical reaction Methods 0.000 description 2
- 238000002372 labelling Methods 0.000 description 2
- ORILYTVJVMAKLC-UHFFFAOYSA-N Adamantane Natural products C1C(C2)CC3CC1CC2C3 ORILYTVJVMAKLC-UHFFFAOYSA-N 0.000 description 1
- 238000007792 addition Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000008569 process Effects 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 238000004904 shortening Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
- G06V10/32—Normalisation of the pattern dimensions
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Computation (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Computational Linguistics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Evolutionary Biology (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Multimedia (AREA)
- Image Analysis (AREA)
Abstract
The invention relates to a clothing attribute identification method based on migration significance prior information, which specifically comprises the following steps: step S1: acquiring image data of a garment image for attribute marking and preprocessing; step S2: inputting the saliency map into a saliency detection network to obtain a saliency map, and overlapping the saliency map with a clothing image to form a clothing image with saliency prior information; step S3: executing steps S1-S2 on each clothing image to obtain all clothing images with significance prior information, and inputting the clothing images into a classification convolution neural network for training until convergence; step S4: and (4) preprocessing the image to be detected in the step S1, then executing the step S2, obtaining the corresponding image to be detected with the significance prior information, inputting the image to be detected into the trained classified convolutional neural network, identifying the clothing attribute, and outputting the clothing attribute in the image to be detected. Compared with the prior art, the method has the advantages of stronger generalization capability, improvement of attribute identification accuracy, reduction of labor cost input and the like.
Description
Technical Field
The invention relates to the field of computer vision and deep learning, in particular to a clothing attribute identification method based on migration significance priori information.
Background
The clothing attribute is the most direct basic information displayed to the consumer by clothing commodities, a matching relation between the consumer and the clothing commodities is constructed, and the consumer is directly guided to purchase. In the past, when a customer goes off a clothing store online, shopping guide personnel often recommend clothing items meeting the customer's needs according to the purchase needs extracted by the customer and then convert the purchase needs into corresponding clothing attributes according to experience, such as style, color, material, accessories and the like. Today, online consumption has become a way of shopping for more and more people, and manual shopping guide is not possible given the hundreds or thousands of shopping needs per second. Therefore, the automatic recommendation of the goods of the e-commerce to potential or target consumers becomes an indispensable function of the e-commerce platform. Particularly for frequently-replaced clothing commodities, automation of clothing commodity attribute identification is realized by utilizing an algorithm in an e-commerce clothing picture, the online shopping guide can be accurately met for the purchasing demand of a consumer, and the online shopping guide has important practical significance for improving the e-commerce profit, shortening the purchasing time and pulling the consumption internal demand.
In the existing clothing attribute identification method, prior information is an important factor influencing the final attribute identification accuracy, and the prior information can directly assist an algorithm to identify clothing in a picture. At present, in the method for identifying the clothing attributes, prior information is divided into two types, one type is prior information of a mark point, the mark point gives a plurality of picture coordinates representing the positions of all parts of clothing in pictures, such as positions of a collar, cuffs and the like; the other type is boundary box prior information, and the boundary box surrounds the position of the clothing in the picture. Although both the prior information can be directly in the image to provide the information of the clothing on the space, the prior information is generated by manual labeling, which is very labor-consuming. By using the two kinds of prior information, the application occasions of the method are limited, the generalization difficulty of the method is increased, and the cost of algorithm design is indirectly increased.
Disclosure of Invention
The invention aims to overcome the defects that the prior information generated by manual marking in the prior art limits the application occasions of the method and increases the generalization difficulty of the method, and provides a clothing attribute identification method based on the prior information with the significance of migration.
The purpose of the invention can be realized by the following technical scheme:
a clothing attribute identification method based on migration significance prior information specifically comprises the following steps:
step S1: acquiring image data of a garment image, performing attribute annotation, and preprocessing the garment image;
step S2: the clothing image is input to a significance detection network for significance prediction to obtain a significance map of the clothing image, and the significance map and the clothing image are overlapped to form a clothing image with significance prior information;
step S3: executing steps S1-S2 on each clothing image to obtain clothing images with significance priori information corresponding to all the clothing images, and inputting the clothing images with the significance priori information corresponding to all the clothing images into a classification convolution neural network for training until the classification convolution neural network converges;
step S4: the image to be detected is preprocessed in the step S1, and then the step S2 is executed to obtain the corresponding image to be detected with the significant prior information, where the image to be detected with the significant prior information is input to the trained classified convolutional neural network to identify the clothing attributes, and the clothing attributes in the image to be detected are output.
The pre-processing of the garment image includes size normalization of the garment image.
Preferably, the size-normalized interpolation algorithm uses bilinear interpolation.
The size normalization corresponds to a garment image size of 256 x 256.
The preprocessing of the clothing image further comprises the enhancing operation of the clothing image, and the enhancing operation comprises horizontal turning and brightness transformation of the clothing image.
The attribute labels include 10 attribute tags.
The number of channels of the clothing image with the significant priori information is 4.
In the step S2, the saliency map is adjusted to have the same size as the input clothing image by bilinear interpolation.
Compared with the prior art, the invention has the following beneficial effects:
1. in the invention, on the selection of the prior information, a significance detection network with mature migration is adopted, and the significance prior information is generated by utilizing the significance of the clothing body in the image, so that the marking of marking points or boundary frames is avoided, the labor cost is reduced, and the recommendation of clothing commodities is effectively realized.
2. The method utilizes the classification convolution neural network to train the known clothing image and then identifies the clothing attribute of the image to be detected, thereby effectively improving the identification accuracy of the clothing attribute.
3. The invention firstly carries out size normalization processing on the clothing image, and enhances the clothing attribute through horizontal turning and brightness conversion, thereby improving the generalization capability of the method.
Drawings
FIG. 1 is a schematic flow diagram of the present invention;
FIG. 2 is a pre-processed image of a garment according to one embodiment of the present invention;
FIG. 3 is an image of a garment undergoing an enhancement operation according to one embodiment of the present invention;
fig. 4 is a schematic structural diagram of a saliency detection network according to an embodiment of the present invention;
FIG. 5 is a saliency map of an image of a garment according to one embodiment of the present invention;
fig. 6 is a schematic structural diagram of a classified convolutional neural network according to an embodiment of the present invention.
Detailed Description
The invention is described in detail below with reference to the figures and specific embodiments. The present embodiment is implemented on the premise of the technical solution of the present invention, and a detailed implementation manner and a specific operation process are given, but the scope of the present invention is not limited to the following embodiments.
As shown in fig. 1, a method for identifying a garment attribute based on migration significance prior information specifically includes the following steps:
step S1: acquiring image data of a garment image, performing attribute annotation, and preprocessing the garment image;
step S2: the clothing image is input to a significance detection network for significance prediction to obtain a significance map of the clothing image, and the significance map and the clothing image are overlapped to form a clothing image with significance prior information;
step S3: executing steps S1-S2 on each garment image to obtain garment images with significance prior information corresponding to all the garment images, and inputting the garment images with significance prior information corresponding to all the garment images into a classification convolution neural network for training until the classification convolution neural network converges;
step S4: when the online identification is carried out, the image to be detected is preprocessed in the step S1, then the step S2 is executed, the image to be detected with the significance prior information corresponding to the image to be detected is obtained, the image to be detected with the significance prior information is input into the trained classified convolutional neural network, the clothing attribute is identified, and the clothing attribute in the image to be detected is output.
The pre-processing of the garment image includes size normalization of the garment image.
Preferably, the size-normalized interpolation algorithm uses bilinear interpolation.
The size normalization corresponds to a garment image size of 256 x 256.
The preprocessing of the clothing image also comprises the enhancement operation of the clothing image, and the enhancement operation comprises horizontal turning and brightness transformation of the clothing image.
The attribute label includes 10 attribute tags.
The number of channels of the garment image with significance prior information is 4.
In step S2, the saliency map is adjusted to have the same size as the input clothing image by bilinear interpolation.
Example one
A clothing attribute identification method based on migration significance prior information specifically comprises the following steps:
step S1: acquiring image data of a garment image, performing attribute annotation, and performing preprocessing and enhancing operations on the garment image;
step S2: the clothing image is input to a significance detection network for significance prediction to obtain a significance map of the clothing image, and the significance map and the clothing image are overlapped to form a clothing image with significance prior information;
step S3: executing steps S1-S2 on each garment image to obtain garment images with significance prior information corresponding to all the garment images, and inputting the garment images with significance prior information corresponding to all the garment images into a classification convolution neural network for training until the classification convolution neural network converges;
step S4: when on-line identification is carried out, preprocessing is carried out on the image to be detected in the step S1, then the step S2 is executed, the image to be detected with the significance prior information corresponding to the image to be detected is obtained, the image to be detected with the significance prior information is input into the trained classified convolutional neural network, clothing attributes are identified, clothing attributes in the image to be detected are output, and the output with the maximum probability is the final prediction result.
Wherein the step S1 specifically includes:
step S101: performing attribute labeling on each clothing image, dividing the label into 10 attributes, and forming a label set Y e (Y)1,y2,...,ynN is attribute total number 10, which is respectively color, collar, color style, sleeve length, zipper, waistband, button, clothing length, clothing type and sleeve style;
step S102: scaling each image to 256 × 256, and completing image size normalization by using bilinear interpolation in a corresponding interpolation algorithm, wherein a specific result is shown in fig. 2;
step S103: and horizontally turning and converting the size-normalized clothing image, wherein the brightness conversion coefficient is +/-0.2, and the specific result is shown in fig. 3.
Step S2 specifically includes:
step S201: as shown in fig. 4, the VGG-16-based significance detection network is composed of five network structures including ConvBlock1, 2, 3, 4 and 5. Each ConvBlock contains two convolutional layer, activation functions, of which ConvBlock1, 2, 3 contains the most valued pooling. After the clothing image is input into the first network structure, the convolution characteristic is extracted and then is used as the input of the next network structure to repeat the extraction process. Convolutive features output by ConvBlock5 were 32 x 32 since the most pooling would reduce the spatial size of the features;
step S202: the preprocessed and enhanced clothing image is used as network input, a saliency map is output, finally, the saliency map is adjusted from 32 x 32 to 256 x 256 by bilinear interpolation, and the final saliency map is shown in fig. 5.
Step S3 specifically includes:
step S301: as shown in fig. 6, the VGG-16-based classification convolutional neural network is adopted, and the classification convolutional neural network is composed of six network structures, namely a convolutional structure and a full connection layer, including ConvBlock1, 2, 3, 4, 5 and FC. Each ConvBlock contains two convolutional layers, an activation function, and a most valued pooling. After the clothing image with the significant prior information passes through five network structures, the extracted convolution features are arranged into a one-dimensional vector to be used as the input of the FC, and each output node of the FC corresponds to the attribute of the clothing. The classification convolutional neural network adopts Softmax as a prediction classifier, a training loss function is cross entropy, an optimizer is ADAM, and the learning rate is 0.01.
Step S302: and superposing the saliency map and the clothing image to generate a clothing image with saliency prior as the input of a network, wherein the number of channels of the input image is 4.
In addition, it should be noted that the specific embodiments described in the present specification may have different names, and the above descriptions in the present specification are only illustrations of the structures of the present invention. Minor or simple variations in the structure, features and principles of the present invention are included within the scope of the present invention. Various modifications or additions may be made to the described embodiments or methods may be similarly employed by those skilled in the art without departing from the scope of the invention as defined in the appending claims.
Claims (8)
1. A clothing attribute identification method based on migration significance prior information is characterized by comprising the following steps:
step S1: acquiring image data of a garment image, performing attribute annotation, and preprocessing the garment image;
step S2: the clothing image is input to a significance detection network for significance prediction to obtain a significance map of the clothing image, and the significance map and the clothing image are overlapped to form a clothing image with significance prior information;
step S3: executing steps S1-S2 on each clothing image to obtain clothing images with significance priori information corresponding to all the clothing images, and inputting the clothing images with the significance priori information corresponding to all the clothing images into a classification convolution neural network for training until the classification convolution neural network converges;
step S4: the image to be detected is preprocessed in the step S1, and then the step S2 is executed to obtain the corresponding image to be detected with the significant prior information, where the image to be detected with the significant prior information is input to the trained classified convolutional neural network to identify the clothing attributes, and the clothing attributes in the image to be detected are output.
2. The method for identifying the clothing attribute based on the migration significance prior information as claimed in claim 1, wherein the preprocessing of the clothing image comprises size normalization of the clothing image.
3. The method for identifying clothing attributes based on the migration significance prior information as claimed in claim 2, wherein the interpolation algorithm of the size normalization adopts bilinear interpolation.
4. The method according to claim 2, wherein the size of the garment image corresponding to the size normalization is 256 × 256.
5. The method for identifying the clothing attribute based on the migration significance prior information as claimed in claim 2, wherein the preprocessing of the clothing image further comprises a clothing image enhancement operation, and the clothing image enhancement operation comprises horizontal flipping and brightness transformation.
6. The method for identifying clothing attributes based on the migration significance prior information is characterized in that the attribute labels comprise 10 attribute labels.
7. The method for identifying the clothing attribute based on the migration significance prior information is characterized in that the number of channels of the clothing image with the significance prior information is 4.
8. The method for identifying clothing attributes based on the migrated significant prior information as claimed in claim 1, wherein the corresponding size of the significant map in step S2 is adjusted to be the same as the size of the input clothing image through bilinear interpolation.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202010043730.5A CN111291784A (en) | 2020-01-15 | 2020-01-15 | Clothing attribute identification method based on migration significance prior information |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202010043730.5A CN111291784A (en) | 2020-01-15 | 2020-01-15 | Clothing attribute identification method based on migration significance prior information |
Publications (1)
Publication Number | Publication Date |
---|---|
CN111291784A true CN111291784A (en) | 2020-06-16 |
Family
ID=71021212
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202010043730.5A Pending CN111291784A (en) | 2020-01-15 | 2020-01-15 | Clothing attribute identification method based on migration significance prior information |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN111291784A (en) |
Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107346436A (en) * | 2017-06-29 | 2017-11-14 | 北京以萨技术股份有限公司 | A kind of vision significance detection method of fused images classification |
CN107967474A (en) * | 2017-11-24 | 2018-04-27 | 上海海事大学 | A kind of sea-surface target conspicuousness detection method based on convolutional neural networks |
CN108334901A (en) * | 2018-01-30 | 2018-07-27 | 福州大学 | A kind of flowers image classification method of the convolutional neural networks of combination salient region |
CN109409384A (en) * | 2018-09-30 | 2019-03-01 | 内蒙古科技大学 | Image-recognizing method, device, medium and equipment based on fine granularity image |
CN109615028A (en) * | 2019-01-22 | 2019-04-12 | 浙江大学 | A kind of medicinal plant classification method based on notable figure |
CN110598610A (en) * | 2019-09-02 | 2019-12-20 | 北京航空航天大学 | Target significance detection method based on neural selection attention |
-
2020
- 2020-01-15 CN CN202010043730.5A patent/CN111291784A/en active Pending
Patent Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107346436A (en) * | 2017-06-29 | 2017-11-14 | 北京以萨技术股份有限公司 | A kind of vision significance detection method of fused images classification |
CN107967474A (en) * | 2017-11-24 | 2018-04-27 | 上海海事大学 | A kind of sea-surface target conspicuousness detection method based on convolutional neural networks |
CN108334901A (en) * | 2018-01-30 | 2018-07-27 | 福州大学 | A kind of flowers image classification method of the convolutional neural networks of combination salient region |
CN109409384A (en) * | 2018-09-30 | 2019-03-01 | 内蒙古科技大学 | Image-recognizing method, device, medium and equipment based on fine granularity image |
CN109615028A (en) * | 2019-01-22 | 2019-04-12 | 浙江大学 | A kind of medicinal plant classification method based on notable figure |
CN110598610A (en) * | 2019-09-02 | 2019-12-20 | 北京航空航天大学 | Target significance detection method based on neural selection attention |
Non-Patent Citations (2)
Title |
---|
ZIWEI LIU ET AL.: "DeepFashion: Powering Robust Clothes Recognition and Retrieval with rich annotations", 《2016 IEEE CONFERENCE ON COMPUTER VISION AND PATTERN RECOGNITION (CVPR)》 * |
沈程秀: "基于显著性先验检测和神经网络的图像分类***的设计与实现", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
US11232324B2 (en) | Methods and apparatus for recommending collocating dress, electronic devices, and storage media | |
TWI559242B (en) | Visual clothing retrieval | |
CN110428428A (en) | A kind of image, semantic dividing method, electronic equipment and readable storage medium storing program for executing | |
CN112529768B (en) | Garment editing and generating method based on generation countermeasure network | |
CN107679960A (en) | A kind of recommendation method of the customized clothing based on image of clothing and label text bimodal content analysis | |
US10769198B1 (en) | Systems and methods for product identification using image analysis from image mask and trained neural network | |
US11847800B2 (en) | Systems and methods for analysis of images of apparel in a clothing subscription platform | |
CN110929665A (en) | Natural scene curve text detection method | |
CN105787490A (en) | Commodity fashion identification method and device based on deep learning | |
CN111784845A (en) | Virtual fitting method and device based on artificial intelligence, server and storage medium | |
CN110647906A (en) | Clothing target detection method based on fast R-CNN method | |
CN111340782B (en) | Image marking method, device and system | |
CN114997956B (en) | Mother and infant product intelligent recommendation system based on big data | |
CN109948702A (en) | A kind of clothes classification and recommended models based on convolutional neural networks | |
Ghodhbani et al. | You can try without visiting: a comprehensive survey on virtually try-on outfits | |
CN112883926A (en) | Identification method and device for table medical images | |
CN111522979A (en) | Picture sorting recommendation method and device, electronic equipment and storage medium | |
CN111222530A (en) | Fine-grained image classification method, system, device and storage medium | |
CN111291784A (en) | Clothing attribute identification method based on migration significance prior information | |
Ou et al. | An Intelligent Recommendation System for Real Estate Commodity. | |
KR20220063601A (en) | Artificial intelligence data set production method using AI-based fashion analysis engine | |
CN112365324A (en) | Commodity picture detection method suitable for E-commerce platform | |
Nair et al. | Outfit recommendation–using image processing | |
CN113298616A (en) | Paired garment compatibility prediction method based on feature fusion | |
Yang et al. | CLDM: A clothing landmark detector based on mask R-CNN |
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 | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20200616 |
|
RJ01 | Rejection of invention patent application after publication |