CN109597907A

CN109597907A - Dress ornament management method and device, electronic equipment, storage medium

Info

Publication number: CN109597907A
Application number: CN201811497168.2A
Authority: CN
Inventors: 张伟; 陈益民; 陈海峰
Original assignee: Shenzhen Sensetime Technology Co Ltd
Current assignee: Shenzhen Sensetime Technology Co Ltd
Priority date: 2017-12-07
Filing date: 2018-12-07
Publication date: 2019-04-09
Also published as: CN109598249B; CN109598249A

Abstract

The embodiment of the present application discloses a kind of dress ornament management method and device, electronic equipment, storage medium, computer program product, wherein, method includes: to carry out dress ornament identification at least two images to be managed in image set to be managed, obtains at least two dress ornament images；Each image to be managed includes at least two dress ornament images, each corresponding dress ornament of the dress ornament image；At least two dress ornament images are handled, at least one image subset is obtained, at least one target dress ornament image in image set to be managed is obtained based at least one image subset；At least one dress ornament attribute that the dress ornament in each target dress ornament image is obtained based on target dress ornament image is reduced the amount of images of Attribute Recognition, improves the processing speed to image set to be managed by carrying out dress ornament Attribute Recognition to target dress ornament image.

Description

Dress ornament management method and device, electronic equipment, storage medium

Technical field

This application involves computer vision technique, especially a kind of dress ornament management method and device, electronic equipment, storage are situated between Matter, computer program product.

Background technique

The development of smart phone and mobile Internet produces the picture and video data of magnanimity, is based on picture and video counts According to computer vision the relevant technologies also obtain and arrived development at full speed, wherein the identification Yu search of object are computer vision skills One of most important business application of art.The deep learning method of big data driving at present is a big hot spot of artificial intelligence field, It is also commonly applied in Computer Vision Task.In fashion world, dress ornament identification and search technique have very big in business Application prospect.With popularizing for intelligent movable mobile phone, there is the picture data of a large number of users in mobile phone photo album, user can be using meter Calculation machine vision technique is detected, identified and is retrieved to dress ornament in photograph album, and album management is used for.

Summary of the invention

A kind of dress ornament administrative skill provided by the embodiments of the present application.

According to the one aspect of the embodiment of the present application, a kind of dress ornament management method for providing, comprising:

Dress ornament identification is carried out at least two images to be managed in image set to be managed, obtains at least two dress ornament figures Picture；Each image to be managed includes at least two dress ornament images, each corresponding dress ornament of the dress ornament image；

At least two dress ornaments image is handled, at least one image subset is obtained, based on it is described at least one Image subset obtains at least one target dress ornament image in the image set to be managed；

At least one dress ornament category of the dress ornament in each target dress ornament image is obtained based on the target dress ornament image Property.

Optionally, in any of the above-described embodiment of the method for the application, at least two in image set to be managed to It manages image and carries out dress ornament identification, obtain at least two dress ornament images, comprising:

To each image to be managed, the image to be managed is handled using first nerves network, obtains institute State at least one dress ornament frame in image to be managed；

At least one described dress ornament frame is divided from the image to be managed and obtains at least one dress ornament image.

Optionally, described to utilize first nerves network to described to be managed in any of the above-described embodiment of the method for the application Image is handled, at least one dress ornament frame in the image to be managed is obtained, comprising:

Feature extraction is carried out to the image to be managed, obtains fisrt feature；

The alternative frame of at least one of the image to be managed is obtained based on the fisrt feature；

At least one dress ornament frame is determined based at least one described alternative frame and the fisrt feature.

Optionally, described based at least one described alternative frame and described in any of the above-described embodiment of the method for the application Fisrt feature determines at least one dress ornament frame, comprising:

The fisrt feature and at least one described alternative frame are handled using pool area layer, acquisition is described at least One dress ornament frame.

Optionally, described based on described in target dress ornament image acquisition in any of the above-described embodiment of the method for the application At least one dress ornament attribute of dress ornament in each target dress ornament image, comprising:

Identifying processing is carried out to the target dress ornament image using nervus opticus network, is obtained in the target dress ornament image Dress ornament at least one dress ornament attribute.

Optionally, in any of the above-described embodiment of the method for the application, the nervus opticus network includes shared branch, at least One segmentation branch and at least one classification branch；

It is described that identifying processing is carried out to the target dress ornament image using nervus opticus network, obtain the target dress ornament figure At least one dress ornament attribute of dress ornament as in, comprising:

The target dress ornament image is handled by the shared branch, obtains second feature；

The second feature is inputted at least one described segmentation branch and at least one described classification branch respectively, is obtained At least one segmentation result and at least one classification results of the dress ornament image；

At least one of the dress ornament is obtained based at least one described classification results and at least one described segmentation result Dress ornament attribute.

Optionally, it in any of the above-described embodiment of the method for the application, is handled at least two dress ornaments image, At least one image subset is obtained, at least one of described image set to be managed is obtained based at least one described image subset After target dress ornament image, further includes:

The quantity for the dress ornament for including in the image set to be managed is determined based on the quantity of the target dress ornament image.

Optionally, in any of the above-described embodiment of the method for the application, it is described at least one described dress ornament image at Reason, is obtained at least one image subset, is obtained in the image set to be managed at least based at least one described image subset One target dress ornament image, comprising:

At least two dress ornaments image is clustered, at least one image subset is obtained；

Each image subset determines a dress ornament image as target respectively from least one described image subset respectively Dress ornament image obtains at least one target dress ornament image；Similarity between the every two target dress ornament image is less than setting Value.

Optionally, described that at least two dress ornaments image is gathered in any of the above-described embodiment of the method for the application Class obtains at least one image subset, comprising:

At least two dress ornaments image is clustered based on the similarity between at least two dress ornaments image, is obtained Obtain at least one image subset.

Optionally, in any of the above-described embodiment of the method for the application, further includes:

At least one matched combined at least one dress ornament attribute based on the dress ornament and collocation database, obtain with At least one matching dress ornament of the clothing matching, and/or at least one match attribute with dress ornament attribute collocation is obtained, The matched combined includes the dress ornament that at least two attributes are mutually matched, and at least two dress ornaments in the matched combined are each other With dress ornament.

Optionally, in any of the above-described embodiment of the method for the application, at least one dress ornament category based on the dress ornament Property and collocation database at least one matched combined, acquisition dress ornament is matched at least one of the clothing matching, comprising:

At least one dress ornament attribute based on the dress ornament searches acquisition in the collocation database has identical or phase Dress ornament is prestored like dress ornament attribute；

The corresponding matching dress ornament of dress ornament is prestored as at least one with the clothing matching using what the lookup obtained With dress ornament.

Optionally, in any of the above-described embodiment of the method for the application, at least one dress ornament category based on the dress ornament Property and collocation database at least one matched combined, obtain with the dress ornament attribute collocation at least one match attribute, Include:

At least one matching dress ornament is obtained based on the corresponding matched combined of dress ornament that prestores obtained of searching；

Using at least one corresponding at least one dress ornament attribute of matching dress ornament of the acquisition as with the dress ornament attribute At least one match attribute of collocation.

Obtain dressing relevant information；

At least one dress ornament attribute based on the dressing relevant information and each dress ornament, from the image to be managed It concentrates screening to obtain at least one and recommends dress ornament or at least a set of recommendation dress ornament combination.

Optionally, in any of the above-described embodiment of the method for the application, the dressing relevant information is comprised at least one of the following: Geographical location information, Weather information, occasion information, chat record information.

Optionally, described based on the dressing relevant information and described every in any of the above-described embodiment of the method for the application At least one dress ornament attribute of a dress ornament, screening obtains at least one recommendation dress ornament from the image set to be managed, comprising:

At least one dress ornament attribute is obtained based on the dressing relevant information, each dressing relevant information is corresponding at least A kind of dress ornament attribute；

At least one of each dress ornament at least one dress ornament attribute and the image set to be managed based on the acquisition Dress ornament attribute is matched, and is obtained at least one and is recommended dress ornament.

Optionally, in any of the above-described embodiment of the method for the application, at least one dress ornament category based on the acquisition Property matched at least one dress ornament attribute of each dress ornament in the image set to be managed, obtain at least one recommend clothes Decorations, comprising:

In response to all clothes that the corresponding dress ornament attribute of the dress ornament includes at least one dress ornament attribute of the acquisition Attribute is adornd, using the dress ornament as the recommendation dress ornament.

Optionally, described based on the dressing relevant information and described every in any of the above-described embodiment of the method for the application At least one dress ornament attribute of a dress ornament, screening obtains at least a set of recommendation dress ornament combination, packet from the image set to be managed It includes:

At least one of each dress ornament at least one dress ornament attribute and the image set to be managed based on the acquisition Dress ornament attribute is matched, and is obtained at least two and is recommended dress ornament；

At least two based on the acquisition recommend dress ornament to obtain at least a set of recommendation dress ornament combination.

Optionally, in any of the above-described embodiment of the method for the application, at least one dress ornament category based on the acquisition Property matched at least one dress ornament attribute of each dress ornament in the image set to be managed, obtain at least two recommend clothes Decorations, comprising:

In response to all clothes that the corresponding dress ornament attribute of the dress ornament includes at least one dress ornament attribute of the acquisition Attribute is adornd, using the dress ornament as the recommendation dress ornament, obtains at least two recommendation dress ornaments.

Optionally, in any of the above-described embodiment of the method for the application, at least two based on the acquisition recommend clothes Decorations obtain at least a set of recommendation dress ornament combination, comprising:

It is to be established between at least two recommendations dress ornament based on the corresponding dress ornament attribute of at least two recommendations dress ornament Matching relationship；

Using at least two recommendation dress ornaments for establishing matching relationship as recommendation dress ornament combination.

According to the another aspect of the embodiment of the present application, a kind of dress ornament managing device for providing, comprising:

Dress ornament recognition unit is obtained for carrying out dress ornament identification at least two images to be managed in image set to be managed Take at least two dress ornament images；Each image to be managed includes at least two dress ornament images, each dress ornament image pair Answer a dress ornament；

Image subset unit obtains at least one image subset for handling at least two dress ornaments image, At least one target dress ornament image in the image set to be managed is obtained based at least one described image subset；

Attribute obtaining unit, for obtaining the dress ornament in each target dress ornament image based on the target dress ornament image At least one dress ornament attribute.

Optionally, in any of the above-described Installation practice of the application, the dress ornament recognition unit, comprising:

Dress ornament frame obtains module, is used for each image to be managed, using first nerves network to described to be managed Image is handled, at least one dress ornament frame in the image to be managed is obtained；

Divide module, obtains at least one clothes for dividing at least one described dress ornament frame from the image to be managed Adorn image.

Optionally, in any of the above-described Installation practice of the application, the segmentation module is specifically used for described to be managed Image carries out feature extraction, obtains fisrt feature；At least one of described image to be managed is obtained based on the fisrt feature Alternative frame；At least one dress ornament frame is determined based at least one described alternative frame and the fisrt feature.

Optionally, in any of the above-described Installation practice of the application, the segmentation module is at least one is standby based on described When frame and the fisrt feature being selected to determine at least one dress ornament frame, for using pool area layer to the fisrt feature and described At least one alternative frame is handled, at least one described dress ornament frame is obtained.

Optionally, in any of the above-described Installation practice of the application, the attribute obtaining unit is specifically used for utilizing second Neural network carries out identifying processing to the target dress ornament image, obtains at least one of the dress ornament in the target dress ornament image Dress ornament attribute.

Optionally, in any of the above-described Installation practice of the application, the nervus opticus network includes shared branch, at least One segmentation branch and at least one classification branch；

The attribute obtaining unit is obtained for being handled by the shared branch the target dress ornament image Second feature；The second feature is inputted at least one described segmentation branch and at least one described classification branch respectively, is obtained Obtain at least one segmentation result and at least one classification results of the dress ornament image；Based at least one described classification results and At least one described segmentation result obtains at least one dress ornament attribute of the dress ornament.

Optionally, in any of the above-described Installation practice of the application, described device further include:

Quantity statistics unit includes for being determined in the image set to be managed based on the quantity of the target dress ornament image Dress ornament quantity.

Optionally, in any of the above-described Installation practice of the application, described image subset unit, be specifically used for it is described extremely Few two dress ornament images are clustered, at least one image subset is obtained；It is each from least one described image subset respectively Image subset determines that a dress ornament image as target dress ornament image, obtains at least one target dress ornament image respectively；It is described every Similarity between two target dress ornament images is less than setting value.

Optionally, in any of the above-described Installation practice of the application, described image subset unit is to described at least two Dress ornament image is clustered, when obtaining at least one image subset, for based on the phase between at least two dress ornaments image At least two dress ornaments image is clustered like degree, obtains at least one image subset.

Acquiring unit is matched, at least one at least one dress ornament attribute and collocation database based on the dress ornament A matched combined, acquisition match dress ornament at least one of the clothing matching, and/or obtain and dress ornament attribute collocation At least one match attribute, the matched combined include the dress ornament that at least two attributes are mutually matched, in the matched combined At least two dress ornaments match dress ornament each other.

Optionally, in any of the above-described Installation practice of the application, the matching acquiring unit, for being based on the dress ornament At least one dress ornament attribute search to obtain in the collocation database there is same or similar dress ornament attribute to prestore dress ornament； Using it is described search obtain prestore the corresponding matching dress ornament of dress ornament as at least one with the clothing matching and match dress ornament.

Optionally, in any of the above-described Installation practice of the application, the matching acquiring unit, for being based on the dress ornament At least one dress ornament attribute search to obtain in the collocation database there is same or similar dress ornament attribute to prestore dress ornament； At least one matching dress ornament is obtained based on the corresponding matched combined of dress ornament that prestores obtained of searching；By the acquisition At least one corresponding at least one dress ornament attribute of matching dress ornament matches category as at least one that the dress ornament attribute is arranged in pairs or groups Property.

Information acquisition unit, for obtaining dressing relevant information；

Dress ornament recommendation unit, at least one dress ornament category based on the dressing relevant information and each dress ornament Property, screening obtains at least one recommendation dress ornament or at least a set of recommendation dress ornament combination from the image set to be managed.

Optionally, in any of the above-described Installation practice of the application, the dressing relevant information is comprised at least one of the following: Geographical location information, Weather information, occasion information, chat record information.

Optionally, in any of the above-described Installation practice of the application, the dress ornament recommendation unit, for being based on the dressing Relevant information obtains at least one dress ornament attribute, each corresponding at least one dress ornament attribute of the dressing relevant information；Based on institute State at least one dress ornament attribute progress of each dress ornament in at least one dress ornament attribute and the image set to be managed of acquisition Match, obtains at least one and recommend dress ornament.

Optionally, in any of the above-described Installation practice of the application, the dress ornament recommendation unit is based on the acquisition At least one dress ornament attribute is matched at least one dress ornament attribute of each dress ornament in the image set to be managed, is obtained extremely When a few recommendation dress ornament, for including at least one dress ornament category of the acquisition in response to the corresponding dress ornament attribute of the dress ornament All dress ornament attributes in property, using the dress ornament as the recommendation dress ornament.

Optionally, in any of the above-described Installation practice of the application, the dress ornament recommendation unit, for being based on the dressing Relevant information obtains at least one dress ornament attribute, each corresponding at least one dress ornament attribute of the dressing relevant information；Based on institute State at least one dress ornament attribute progress of each dress ornament in at least one dress ornament attribute and the image set to be managed of acquisition Match, obtains at least two and recommend dress ornament；At least two based on the acquisition recommend dress ornament to obtain at least a set of recommendation dress ornament group It closes.

Optionally, in any of the above-described Installation practice of the application, the dress ornament recommendation unit is based on the acquisition At least one dress ornament attribute is matched at least one dress ornament attribute of each dress ornament in the image set to be managed, is obtained extremely When few two recommendation dress ornaments, for including at least one dress ornament category of the acquisition in response to the corresponding dress ornament attribute of the dress ornament Property in all dress ornament attributes obtain at least two recommendation dress ornaments using the dress ornament as the recommendation dress ornament.

Optionally, in any of the above-described Installation practice of the application, the dress ornament recommendation unit is based on the acquisition When at least two recommendation dress ornaments obtain at least a set of recommendation dress ornament combination, for corresponding based on at least two recommendations dress ornament Dress ornament attribute is to establish matching relationship between at least two recommendations dress ornament；At least two of matching relationship is established by described Recommend dress ornament as recommendation dress ornament combination.

According to the another aspect of the embodiment of the present application, a kind of electronic equipment provided, including processor, the processor Including dress ornament managing device described in any one as above.

According to the still another aspect of the embodiment of the present application, a kind of electronic equipment that provides, comprising: memory, for storing Executable instruction；

And processor, it is as above any one to complete that the executable instruction is executed for communicating with the memory The operation of the item dress ornament management method.

According to another aspect of the embodiment of the present application, a kind of computer storage medium provided, for storing computer The instruction that can be read, which is characterized in that described instruction is performed the behaviour for executing dress ornament management method described in any one as above Make.

According to the other side of the embodiment of the present application, a kind of computer program product provided, including it is computer-readable Code, which is characterized in that when the computer-readable code is run in equipment, the processor execution in the equipment is used for Realize the instruction of dress ornament management method described in any one as above.

A kind of dress ornament management method and device, electronic equipment, storage medium provided based on the above embodiments of the present application is right At least two images to be managed in image set to be managed carry out dress ornament identification, obtain at least two dress ornament images；Each to pipe Managing image includes at least two dress ornament images, each corresponding dress ornament of the dress ornament image；To at least two dress ornament images into Row processing, obtains at least one image subset, obtains at least one of image set to be managed based at least one image subset Target dress ornament image；At least one dress ornament attribute of the dress ornament in each target dress ornament image is obtained based on target dress ornament image, By carrying out dress ornament Attribute Recognition to target dress ornament image, reduces the amount of images of Attribute Recognition, improve to figure to be managed The processing speed of image set.

Below by drawings and examples, the technical solution of the application is described in further detail.

Detailed description of the invention

The attached drawing for constituting part of specification describes embodiments herein, and together with description for explaining The principle of the application.

The application can be more clearly understood according to following detailed description referring to attached drawing, in which:

Fig. 1 is a flow diagram of dress ornament management method provided by the embodiments of the present application.

Fig. 2 is another flow diagram of dress ornament management method provided by the embodiments of the present application.

Fig. 3 is the process signal that dress ornament frame is obtained in another embodiment of dress ornament management method provided by the embodiments of the present application Figure.

Fig. 4 is the structural representation of first nerves network in an example of dress ornament management method provided by the embodiments of the present application Figure.

Fig. 5 is a structural schematic diagram of nervus opticus network in dress ornament management method provided by the embodiments of the present application.

Fig. 6 is to obtain the one of dress ornament attribute using nervus opticus network in dress ornament management method provided by the embodiments of the present application A flow diagram.

Fig. 7 is another structural schematic diagram of nervus opticus network in dress ornament management method provided by the embodiments of the present application.

Fig. 8 is the flow diagram of the another embodiment of dress ornament management method provided by the embodiments of the present application.

Fig. 9 is multiple images in image set to be managed in one example of dress ornament management method provided by the embodiments of the present application Schematic diagram.

Figure 10 is an also flow diagram for dress ornament management method provided by the embodiments of the present application.

Figure 11 be dress ornament management method provided by the embodiments of the present application another example in searched in the database according to attribute The schematic diagram of rope.

Figure 12 is another flow diagram of dress ornament management method provided by the embodiments of the present application.

Figure 13 is that one that recommends dress ornament is obtained in the another embodiment of dress ornament management method provided by the embodiments of the present application Flow diagram.

Figure 14 is to obtain to recommend the another of dress ornament in the another embodiment of dress ornament management method provided by the embodiments of the present application Flow diagram.

Figure 15 is a structural schematic diagram of dress ornament managing device provided by the embodiments of the present application.

Figure 16 is the structural representation suitable for the electronic equipment of the terminal device or server that are used to realize the embodiment of the present application Figure.

Specific embodiment

The various exemplary embodiments of the application are described in detail now with reference to attached drawing.It should also be noted that unless in addition having Body explanation, the unlimited system of component and the positioned opposite of step, numerical expression and the numerical value otherwise illustrated in these embodiments is originally The range of application.

Simultaneously, it should be appreciated that for ease of description, the size of various pieces shown in attached drawing is not according to reality Proportionate relationship draw.

Be to the description only actually of at least one exemplary embodiment below it is illustrative, never as to the application And its application or any restrictions used.

Technology, method and apparatus known to person of ordinary skill in the relevant may be not discussed in detail, but suitable In the case of, the technology, method and apparatus should be considered as part of specification.

It should also be noted that similar label and letter indicate similar terms in following attached drawing, therefore, once a certain Xiang Yi It is defined in a attached drawing, then in subsequent attached drawing does not need that it is further discussed.

Fig. 1 is a flow diagram of dress ornament management method provided by the embodiments of the present application.This method can be by any Electronic equipment executes, such as terminal device, server, mobile device etc..

Step 110, dress ornament identification is carried out at least two images to be managed in image set to be managed, obtains at least two Dress ornament image.

Wherein, each image to be managed includes at least two dress ornament images, the corresponding dress ornament of each dress ornament image；It is optional Ground, image set to be managed can be the image collection that mobile phone photo album of user etc. includes multiple dress ornament images.

It is alternatively possible to dress ornament image recognition is carried out using deep learning network handles management image, and to be managed Before image carries out dress ornament image recognition, the mass picture training deep learning network for being labelled with dress ornament position frame, training are used Dress ornament detection and identification model are compared between clothes between feature vector using feature vector obtained in dress ornament identification model Distance, the dress ornament in a large amount of pictures is clustered, these feature vectors can be used for the dress ornament picture in database It scans for, obtains similar or with money dress ornament.The method that the application proposes is not limited to the depth convolution detection algorithm of mainstream, also Including Faster RCNN, SSD, YOLO etc.；Dress ornament recognizer be not limited to mainstream for depth convolutional Neural sorter network, It further include AlexNet, VGG, Inception v1, v2, v3, v4, ResNet, ResNeXt etc.；For clustering and being searched with money dress ornament The feature vector of rope is eventually used for the full articulamentum feature of classification for sorter network, but can also be learnt using network Otherness between different money clothes obtains the higher feature of identification, improves the precision of retrieval.

Step 120, at least two dress ornament images are handled, obtains at least one image subset, be based at least one Image subset obtains at least one target dress ornament image in image set to be managed.

Wherein, each image subset may include at least one dress ornament image.

Optionally, to the executable cluster of at least two dress ornament images, classification or by the processing such as neural network classification, make phase It is gathered in an image subset like higher dress ornament image is spent, it is believed that the dress ornament image in each image subset corresponds to same Dress ornament, for example, user, which wears same dress, has taken 5 photos, when obtaining dress ornament attribute, first by the clothes from photo In identify, obtain the dress ornament image of corresponding clothes, clustering processings then carried out to 5 dress ornament images, 5 dress ornament images are gathered Collect in same image subset, a dress ornament image is obtained from image subset can be used as target dress ornament image.Above-mentioned mistake Journey reduces the quantity (an only corresponding target dress ornament image in each image subset) for needing the image of recognition property, improves The speed of dress ornament Attribute Recognition.

Step 130, at least one dress ornament attribute of the dress ornament in target dress ornament image is obtained based on target dress ornament image.

In the embodiment of the present application, using target dress ornament image as the basis for obtaining dress ornament attribute, in target dress ornament image Do not include other background contents other than dress ornament, reduces interference of the background content to dress ornament determined property, improve clothes Adorn the accuracy of Attribute Recognition.

Optionally, target dress ornament image is handled using one or more neural networks, obtains at least the one of dress ornament Kind dress ornament attribute, wherein neural network can be the deep neural network of any achievable dress ornament Attribute Recognition, using one Neural network obtains at least one dress ornament attribute, or obtains a kind of clothes respectively using each neural network in multiple neural networks Attribute is adornd, the embodiment of the present application does not limit the specific structure and quantity of neural network.Dress ornament attribute is being obtained using neural network Before, the great amount of images for being labelled with dress ornament attribute can be used to be trained the neural network.

The dress ornament management method provided based on the above embodiment, at least two images to be managed in image set to be managed Dress ornament identification is carried out, at least two dress ornament images are obtained；Each image to be managed includes at least two dress ornament images, each described Dress ornament image corresponds to a dress ornament；At least two dress ornament images are handled, at least one image subset is obtained, based at least One image subset obtains at least one target dress ornament image in image set to be managed；It is obtained based on target dress ornament image each At least one dress ornament attribute of dress ornament in target dress ornament image is subtracted by carrying out dress ornament Attribute Recognition to target dress ornament image The amount of images for having lacked Attribute Recognition improves the processing speed to image set to be managed.

Fig. 2 is another flow diagram of dress ornament management method provided by the embodiments of the present application.As shown in Fig. 2, the implementation Example method include:

Step 210, to each image to be managed, handled using first nerves network handles management image, obtain to Manage at least one dress ornament frame in image.

Optionally, it before being identified using first nerves network to dress ornament image, needs to the convolutional neural networks It is trained, training convolutional neural networks realize dress ornament detection, and training data is sample dress ornament picture, these sample dress ornament pictures All it is labelled with the position of dress ornament；When first nerves training, whole network is trained using dress ornament callout box position；When test, Picture is inputted, the coordinate of dress ornament frame is exported, determines the dress ornament frame in image to be managed.

Step 220, at least one dress ornament frame is divided from image to be managed and obtains at least one dress ornament image.

Due to may include multiple dress ornaments in an image to be managed, i.e., for multiple dress ornament frames, in order to each dress ornament Attribute Recognition is carried out, needs respectively to handle the corresponding image of each dress ornament, the embodiment of the present application is by from figure to be managed The segmentation of dress ornament frame is obtained into corresponding dress ornament image as in, using dress ornament image as the basis of processing, it is alternatively possible to segmentation The dress ornament frame of acquisition zooms in and out processing, obtains the identical dress ornament image of multiple sizes, is based on consolidated network pair in order to subsequent Dress ornament image carries out Attribute Recognition processing.

Step 230, at least two dress ornament images are handled, obtains at least one image subset, be based at least one Image subset obtains at least one target dress ornament image in image set to be managed.

At least two dress ornament images obtained to segmentation can carry out the processing such as clustering, to obtain and dress ornament quantity same number Image subset, the corresponding dress ornament of each image subset optionally randomly selects a dress ornament figure from each image subset As being used as target dress ornament image, or takes picture quality in image subset (picture quality includes but is not limited to: clarity, has size Imitate pixel ratio etc.) for preferably dress ornament image as target dress ornament image, the embodiment of the present application, which does not limit, obtains target dress ornament figure The concrete mode of picture.

Step 240, at least one dress ornament attribute of the dress ornament in target dress ornament image is obtained based on target dress ornament image.

By carrying out Attribute Recognition to target dress ornament image, it can be achieved that dress ornament category to all dress ornaments in image set to be processed Property identification, reduce the quantity of dress ornament Attribute Recognition, improve recognition efficiency.

The embodiment of the present application obtains the dress ornament frame in image to be managed by trained first nerves network, realizes More accurate dress ornament position identification, improves the accuracy of dress ornament image acquisition.

Fig. 3 is the process signal that dress ornament frame is obtained in another embodiment of dress ornament management method provided by the embodiments of the present application Figure.As shown in figure 3, step 210 includes: in above-mentioned another embodiment

Step 302, feature extraction is carried out to image to be managed, obtains fisrt feature.

Optionally, the first nerves network in the embodiment of the present application is the network of any achievable dress ornament detection, for example, with Illustrating detection algorithm process for the detection algorithm of Faster Rcnn, Faster Rcnn is two step object detection algorithms, point It Wei not alternative frame generating portion and final frame detection part.

Step 304, the alternative frame of at least one of image to be managed is obtained based on fisrt feature.

The present embodiment realization handles fisrt feature, and obtaining all in image to be managed may include the alternative of dress ornament Frame, there may be inclusion relation, cross references etc. between the alternative frame of acquisition can be obtained by the processing to alternative frame The dress ornament frame of corresponding each dress ornament.

Step 306, at least one dress ornament frame is determined based at least one alternative frame and fisrt feature.

Optionally, fisrt feature and at least one alternative frame are handled using pool area layer, obtains at least one Dress ornament frame.

In the present embodiment, it obtains the network of alternative frame and determines the network share subnetwork layer of dress ornament frame, share portion Fisrt feature is separately won to obtain, for example, shared is partially at least one layer of convolutional layer, it is in the embodiment of the present application that the last layer convolutional layer is defeated Feature out generates at least one alternative frame that network exports as fisrt feature, in conjunction with fisrt feature and alternative frame, by area The processing of domain pond layer (for example, area-of-interest pond layer ROI Pooling) and full articulamentum, from least one alternative frame It selects at least one alternative frame as dress ornament frame, that is, at least one dress ornament frame has been determined.

Fig. 4 is the structural representation of first nerves network in an example of dress ornament management method provided by the embodiments of the present application Figure.As shown in Figure 4, ((corresponding step 306) shares portion to corresponding step 304) to alternative frame generation network with final frame detection network Dividing convolutional neural networks, (such as convolutional layer 1~5, corresponding step 302), alternative frame generate network portion and generate alternative frame, are dress ornament High probability existing for position, final frame detection network refines alternative frame, obtains final dress ornament frame position.

In one or more optional embodiments, dress ornament attribute can include but is not limited to following at least one: dress ornament Classification (e.g., T-shirt, Polo shirt, western clothes etc.), dress ornament wear human nature not (e.g., male, women, neutrality etc.), dress ornament style (e.g., Europe Beauty, Japan and Korea S, rural area etc.), dress ornament texture (e.g., striped, far point, leopard line etc.), dress ornament neckline (e.g., stand-up collar, crew neck, even cap neck Deng), dress ornament wear suitable scene (e.g., date, work, movement etc.), dress ornament cuff (e.g., long sleeves, cotta, 7 points sleeve etc.), dress ornament Color percentage, dress ornament material (e.g., pure cotton, corium, wool etc.), dress ornament wears effect (e.g., close-fitting, loose etc.), dress ornament are long Short (e.g., surplus, shortage of money, middle surplus etc.), garment beauty, dress ornament waist type (e.g., low waist, middle waist, high waist etc.), skirt type.

In one or more optional embodiments, step 130 or step 240 in above-described embodiment include:

Identifying processing is carried out to target dress ornament image using nervus opticus network, obtains the dress ornament in target dress ornament image At least one dress ornament attribute.

The method of identification dress ornament attribute provided by the embodiments of the present application can use nervus opticus network implementations, second mind It has selected layer different in network to be identified for different attributes when implementing through network, has improved Attribute Recognition essence Degree, by taking color as an example, for the identification color attribute more refined, optionally, structure as shown in figure 5, Fig. 5 is that the application is real One structural schematic diagram of nervus opticus network in the dress ornament management method of example offer is provided.

Optionally, nervus opticus network includes shared branch, at least one segmentation branch and at least one classification branch；

Fig. 6 is to obtain the one of dress ornament attribute using nervus opticus network in dress ornament management method provided by the embodiments of the present application A flow diagram.In this embodiment, nervus opticus network may include shared branch, at least one segmentation branch and at least One classification branch, as shown in fig. 6, obtaining the process of dress ornament attribute in the present embodiment can include:

Step 602, target dress ornament image is handled by shared branch, obtains second feature.

The embodiment of the present application is in order to improve Attribute Recognition speed, using the network layer shared to each branch as shared branch, In attribute recognition process, a feature extraction is only carried out to target dress ornament image based on shared branch, it is subsequent based on spy Sign is extracted the second feature obtained and is executed, and optionally, sharing branch includes at least one layer of convolutional layer (e.g., convolutional layer 1-5).

Step 604, second feature is separately input into a few segmentation branch and at least one classification branch, obtains dress ornament At least one segmentation result and at least one classification results of image.

In order to improve the accuracy of dress ornament Attribute Recognition, the embodiment of the present application increases segmentation point on the basis of classification Branch, the accuracy of classification results can be improved in conjunction with segmentation result.

Step 606, at least one dress ornament of dress ornament is obtained based at least one classification results and at least one segmentation result Attribute.

Optionally, dress ornament segmentation network (the segmentation branch in the embodiment of the present application) can be added in nervus opticus network, right Dress ornament carries out the positioning of pixel (pixel) rank, and as shown in Figure 7, Fig. 7 is dress ornament management method provided by the embodiments of the present application Another structural schematic diagram of middle nervus opticus network.In nervus opticus network, segmentation network is added, is partitioned into dress ornament and then counts The rgb value for calculating pixel rank, is mapped in colour system.For example, as shown in Figure 7 using a segmentation branch to dress ornament Color in image is split, and obtains the segmentation result based on color, which is integrated in classification branch, More accurate classification results can be obtained.

In addition, the region of clothes in picture can be extracted according to dress ornament segmentation result, it is sent into another convolutional Neural It is trained in network, the interference of background in picture can be removed in this way, improve dress ornament Attribute Recognition precision.

In one or more optional embodiments, the embodiment of the present application method after step 120, is also based on mesh The quantity of mark dress ornament image determines the quantity for the dress ornament for including in image set to be managed.

The embodiment of the present application realizes each mesh of acquisition due to being clustered the corresponding dress ornament image of identical dress ornament Mark dress ornament image all corresponds to different dress ornaments, is that can determine in image set to be managed to wrap by obtaining target dress ornament image therefore The quantity of all unduplicated dress ornaments included.

Fig. 8 is the flow diagram of the another embodiment of dress ornament management method provided by the embodiments of the present application.Such as Fig. 8 institute Show, the embodiment of the present application method includes:

Step 810, dress ornament identification is carried out at least two images to be managed in image set to be managed, obtains at least two Dress ornament image.

Step 810 is similar with the step 110 of above-described embodiment in the embodiment of the present application, can refer to above-described embodiment to the step Suddenly understood, details are not described herein.

Step 820, at least two dress ornament images are clustered, obtains at least one image subset.

Optionally, at least two dress ornament images can be clustered based on the similarity between at least two dress ornament images, Obtain at least one image subset.By the biggish dress ornament image collection of similarity into an image subset, image two-by-two is determined Between similarity, can based on the distance between corresponding feature of two images (such as: Euclidean distance, COS distance) determine, Or directly at least two dress ornament images are clustered by trained neural network, obtain at least one image subset.

When the similarity between two images reaches setting value, it is believed that the dress ornament in two dress ornament images is same Part；Specifically, may include: to carry out feature extraction respectively at least two dress ornament images in image set to be managed, the is obtained Three features；The distance between at least two third feature based on acquisition determine the distance between at least two dress ornament images；It will At least two dress ornament images that distance is less than or equal to set distance determine that into same cluster, obtaining includes at least one cluster Image set to be managed.The similarity between dress ornament is determined except through the distance between dress ornament image, is also based on dress ornament The similar dress ornament of attribute selection, for example, using the identical dress ornament image of attribute beyond setting quantity in dress ornament attribute as doubtful phase Same dress ornament, then (can be based on the distance between dress ornament image or other modes), which determine doubtful phase, to be judged to doubtful identical dress ornament It is whether identical with dress ornament, using the identical dress ornament image of determination as the dress ornament image of the same dress ornament.

Step 830, each image subset determines a dress ornament image conduct respectively from least one image subset respectively Target dress ornament image obtains at least one target dress ornament image.

Wherein, the similarity between every two target dress ornament image is less than setting value.After completing cluster, each image Concentrate include the same dress ornament of multiple correspondences dress ornament image, at this point, in order to determine the corresponding dress ornament of each image subset, from each A dress ornament image (can be the dress ornament image etc. for extracting or taking at random display dress ornament attribute most) is taken in image subset, Using the dress ornament image as target dress ornament image.Determine a dress ornament image as target dress ornament figure from each image subset Picture can not limit acquisition target by extracting or extracting modes, the embodiment of the present application such as the preferable dress ornament image of quality at random The concrete mode of dress ornament image.

Step 840, at least one dress ornament attribute of the dress ornament in target dress ornament image is obtained based on target dress ornament image.

Step 840 is similar with the step 130 of above-described embodiment or step 240 in the embodiment of the present application, can refer to above-mentioned reality It applies example and details are not described herein is understood to the step.

User may wear in the different pictures of photograph album same dress, and Fig. 9 is clothes provided by the embodiments of the present application Adorn multiple image schematic diagrames in one example of management method in image set to be managed.As shown in Figure 9, user wears same part Clothes has taken 4 photos, and dress ornament cluster is clustered out inside user's photograph album with money dress ornament, can accurately list user in this way Possess which dress ornament and its attribute, cluster is mainly the feature extracted in convolutional neural networks according to dress ornament, according to spy The distance between sign determines whether dress ornament belongs to same dress, as shown in figure 9, the clothes of red or blue box aggregation Distance between the features is smaller, can be brought together.

It it is frequently present of user in album picture wears same part dress ornament and occur in different pictures, in user's dress ornament management In, need to identify that these dress ornaments are same part, and save the part dress ornament information in the database for user.Net is identified in dress ornament In network, all dress ornaments in all pictures can be carried out using the feature (such as full articulamentum 2) of the full articulamentum of the last layer Cluster, obtains the attribute of user's practical dress ornament number possessed and they, in addition, if constructing a biggish dress ornament data Library or the picture database for accessing electric business can also search for similar money or same money in database again according to these features.

The embodiment of the present application can automatically detect and identify the dress ornament in image set to be managed (e.g., user's photograph album etc.) simultaneously It is clustered, the hobby of user can be obtained by excavating the information of these dress ornaments according to cluster, searched in electric quotient data library And recommend suitable dress ornament to user.

Figure 10 is an also flow diagram for dress ornament management method provided by the embodiments of the present application.As shown in Figure 10, the reality Applying a method includes:

Step 1010, dress ornament identification is carried out at least two images to be managed in image set to be managed, obtains at least two A dress ornament image.

Wherein, each image to be managed includes at least two dress ornament images, the corresponding dress ornament of each dress ornament image；It is optional Ground, image set to be managed can be the image collection that mobile phone photo album of user etc. includes multiple dress ornament images.It can be by any The neural network or other dress ornament identification methods that dress ornament identification can be achieved obtain the dress ornament image in image to be managed, and the application is real Apply the mode that the unlimited fixture body of example obtains dress ornament image.

Step 1020, at least two dress ornament images are handled, obtains at least one image subset, be based at least one Image subset obtains at least one target dress ornament image in image set to be managed.

Wherein, each image subset may include at least one dress ornament image.By handling and being obtained based on image subset Target dress ornament image, reduces the amount of images that need to be identified, reduces resource consumption, saves the time of Attribute Recognition.

Step 1030, at least one dress ornament attribute of the dress ornament in target dress ornament image is obtained based on target dress ornament image.

Optionally, a variety of dress ornament attributes can be obtained by an Attribute Recognition network, or passes through multiple Attribute Recognition networks Every kind of dress ornament attribute is obtained respectively, for example, as known using nervus opticus network to target dress ornament image in above-described embodiment Other places reason, obtains at least one dress ornament attribute of dress ornament.

Step 1040, at least one matched combined at least one dress ornament attribute based on dress ornament and collocation database, Acquisition matches dress ornament at least one of clothing matching, and/or obtains at least one match attribute with the collocation of dress ornament attribute.

Wherein, matched combined includes the dress ornament that at least two attributes are mutually matched, at least two dress ornaments in matched combined Dress ornament is matched each other.The embodiment of the present application is dress ornament image with the matching relationship between dress ornament attribute by identification dress ornament attribute In clothing matching match dress ornament accordingly, save the process of labor mix and screening, realize the clothing matching of fast accurate Recommend.

The matched combined that the embodiment of the present application obtains includes the dress ornament that at least two attributes are mutually matched, in conjunction with collocation data Library is that dress ornament recommends the matching dress ornament to match, realizes the automatic recommendation of matching dress ornament, realizes quick clothing matching and push away It recommends.

In one or more optional embodiments, step 1040 includes:

At least one dress ornament attribute based on dress ornament, which is searched to obtain in collocation database, has same or similar dress ornament category Property prestores dress ornament；

Using search obtain prestore the corresponding matching dress ornament of dress ornament as at least one with clothing matching and match dress ornament.

In the embodiment of the present application, by combine it is known collocation database (for example, on internet magnanimity fashion clothing figure Piece information), the dress ornament information that can be arranged in pairs or groups is excavated, for example, currently popular garment coordination etc., knows in conjunction with from dress ornament image The dress ornament attribute being clipped to recommends clothing matching for user, and matching method can include but is not limited to: a. is automatic in fashion picture It identifies the clothes attribute of collocation, and constructs clothes attribute collocation database, searched in the database using the clothes attribute of user Rope, it is automatic to recommend collocation clothes；Wherein, the process of picture search may include: to carry out dress ornament identification, identification to the picture of input Obtained feature vector scans in the database, returns to search result according to the matching of feature vector, can be to the institute of user There is dress ornament to extract feature and store, in user search, similar dress ornament can be searched for according to these features, in addition, can also To store the attribute of dress ornament, user can also scan in the database according to attribute, and Figure 11 mentions for the embodiment of the present application The schematic diagram searched in the database in another example of the dress ornament management method of confession according to attribute.As shown in Figure 11, user Picture can be inputted to be retrieved, text can also be inputted, obtain search result such as " pink down jackets ".B. depth is used Arranging effect in neural convolutional network study fashion picture automatically generates suitable collocation after given user's garment image Clothes.

In one or more optional embodiments, step 1040 includes:

At least one matching dress ornament is obtained based on the corresponding matched combined of dress ornament that prestores obtained is searched；

At least one corresponding at least one dress ornament attribute of matching dress ornament of acquisition is arranged in pairs or groups extremely as with dress ornament attribute A kind of few match attribute.

The embodiment of the present application is obtained the corresponding matching of dress ornament attribute and is belonged to by the collocation database of known matched combined Property, by these match attributes, the Matching Relation of more depth can be obtained to the collocation between dress ornament, for example, between color Collocation etc..

Figure 12 is another flow diagram of dress ornament management method provided by the embodiments of the present application.This method can be by any Electronic equipment executes, such as terminal device, server, mobile device etc..As shown in figure 12, which includes:

Step 1210, dress ornament identification is carried out at least two images to be managed in image set to be managed, obtains at least two A dress ornament image.

Step 1220, at least two dress ornament images are handled, obtains at least one image subset, be based at least one Image subset obtains at least one target dress ornament image in image set to be managed.

Step 1230, at least one dress ornament attribute of the dress ornament in target dress ornament image is obtained based on target dress ornament image.

It does not include removing in dress ornament image using dress ornament image as the basis for obtaining dress ornament attribute in the embodiment of the present application Other background contents other than dress ornament, reduce interference of the background content to dress ornament determined property, improve the knowledge of dress ornament attribute Other accuracy.

Step 1240, dressing relevant information is obtained.

Optionally, dressing relevant information includes but is not limited to following at least one: geographical location information, Weather information, field Close information, chat record information etc..Corresponding recommendation is carried out for different dressing relevant informations, such as: geographical location information For the higher region of temperature, the clothes of salubrious class, such as T-shirt, skirt can be worn with recommended user；Weather information indicates cold When, down jackets, cotton dress etc. can be worn with recommended user；It is obtained by geographical location by big data analysis or data mining In the region, the clothes of prevalence are that user recommends, and job site and the property of user can be known by specifically positioning, Recommend leisure or the dress of formal clothes for user；By chat record information, the place that user will go to can be known, Such as dining room, seashore can recommend suitable clothes to user according to these scenes.

Step 1250, at least one dress ornament attribute based on dressing relevant information He at least one dress ornament, from figure to be managed Screening obtains at least a set of recommendation dress ornament in image set or at least one recommends dress ornament combination.

Influence people dressing factor there are many kinds of, such as: weather, occasion, place etc., the embodiment of the present application by these Influence factor is recommended suitable dress ornament combination for user, is believed with meeting user in different dressing correlations as dressing relevant information Properly focus in the case of breath (for example, different occasions etc.).For example, recommending more refrigerant when acquisition local temperature is higher Dress ornament；It is workplace when obtaining the occasion for needing to attend, formal clothes dress for being suitble to work, etc. will be recommended.This Shen Please embodiment realize intelligent recommendation and neural metwork training it is irrelevant, but extract clothes attribute for recommend have it is very big Use, such as clothes scene, style, classification etc., different dressing relevant informations can correspond to different dress ornament attributes.

The data of dress ornament management method combination internet mass provided by the embodiments of the present application are analyzed, and can be provided for user Suitable clothing matching is used for different social scenes.Further, it is also possible to be combined with the other information in smart phone, as Position, user's friend information etc. are managed, it is intelligent to provide clothing matching recommendation for user.

Figure 13 is that one that recommends dress ornament is obtained in the another embodiment of dress ornament management method provided by the embodiments of the present application Flow diagram.As shown in figure 13, the step 1250 of above-described embodiment includes:

Step 1302, at least one dress ornament attribute is obtained based on dressing relevant information.

Wherein, the corresponding at least one dress ornament attribute of each dressing relevant information, can get current by dressing relevant information Suitable at least one dress ornament attribute.

Step 1304, at least one dress ornament attribute of at least one dress ornament attribute based on acquisition and at least one dress ornament into Row matching obtains at least one and recommends dress ornament.

Optionally, in response to the corresponding dress ornament attribute of dress ornament include obtain at least one dress ornament attribute in all dress ornaments Attribute, using dress ornament as recommendation dress ornament.

For recommending the acquisition of dress ornament, after determinings a dressing relevant information corresponding dress ornament attribute, recommendation dress ornament Including but not limited to whole acquisition dress ornament attributes is needed, the dress ornament for all dress ornament attributes for only including could be suitble to pretend Situation corresponding to information is closed, the probability of error is reduced.

The embodiment of the present application is user according to the recommendation of the dressing relevant information of acquisition by obtaining at least one to recommend dress ornament At least one suitable recommends dress ornament, and optionally, each recommendation dress ornament can be single in the recommendation dress ornament that the embodiment of the present application is recommended Solely dress recommends sandy beach one-piece dress for example, being seashore according to place.

Figure 14 is to obtain to recommend the another of dress ornament in the another embodiment of dress ornament management method provided by the embodiments of the present application Flow diagram.As shown in figure 14, the step 1250 of above-described embodiment includes:

Step 1402, at least one dress ornament attribute is obtained based on dressing relevant information.

Step 1404, at least one dress ornament attribute of at least one dress ornament attribute based on acquisition and at least one dress ornament into Row matching obtains at least two and recommends dress ornament.

The application uses to be more convenient user, and the dress ornament of recommendation is to realize complete dressing, when the recommendation dress ornament of acquisition is When at least two recommendation dress ornaments, user is recommended again after multiple recommendation dress ornaments can be arranged in pairs or groups, for example, when the recommendation obtained When dress ornament is jacket and trousers, jacket and trousers are matched and recommend user, more saves the time of user's collocation.

Step 1406, at least two based on acquisition recommend dress ornament to obtain at least a set of recommendation dress ornament combination.

The dress ornament for obtaining and there are these dress ornament attributes is searched according to the dress ornament attribute of acquisition, is taken these dress ornaments as recommendation Decorations, and recommend dress ornament combination based on recommending dress ornament to obtain, suitable a variety of dressings selection is provided for user, saves a large amount of sieves Select the time.

Optionally, step 1406 includes:

It is to establish matching relationship between at least two recommendation dress ornaments based on the corresponding dress ornament attribute of at least two recommendation dress ornaments；

Recommend dress ornament as recommendation dress ornament combination for establish matching relationship at least two.

In the embodiment of the present application, matching relationship will be established according to its corresponding dress ornament attribute between multiple recommendation dress ornaments, that is, It will arrange in pairs or groups between multiple recommendation dress ornaments, for example, jacket and trousers are arranged in pairs or groups, tippet and longuette arranged in pairs or groups, etc. Deng.It may include at least two recommendation dress ornaments in this matching relationship, establish after matching relationship, be directly in by recommendation dress ornament combination User is now given, is selected for user, saves the time that user voluntarily arranges in pairs or groups.

Those of ordinary skill in the art will appreciate that: realize that all or part of the steps of above method embodiment can pass through The relevant hardware of program instruction is completed, and program above-mentioned can be stored in a computer readable storage medium, the program When being executed, step including the steps of the foregoing method embodiments is executed；And storage medium above-mentioned includes: ROM, RAM, magnetic disk or light The various media that can store program code such as disk.

Figure 15 is a structural schematic diagram of dress ornament managing device provided by the embodiments of the present application.The device of the embodiment can For realizing the above-mentioned each method embodiment of the application.As shown in figure 14, the device of the embodiment includes:

Dress ornament recognition unit 1501, for carrying out dress ornament knowledge at least two images to be managed in image set to be managed Not, at least two dress ornament images are obtained.

Image subset unit 1502 obtains at least one image subset for handling at least two dress ornament images, At least one target dress ornament image in image set to be managed is obtained based at least one image subset.

Attribute obtaining unit 1503, for obtaining the dress ornament in each target dress ornament image based on target dress ornament image extremely A kind of few dress ornament attribute.

The dress ornament managing device provided based on the above embodiment, by carrying out dress ornament Attribute Recognition to target dress ornament image, The amount of images for reducing Attribute Recognition improves the processing speed to image set to be managed.

In one or more optional embodiments, dress ornament recognition unit 1501, comprising:

Dress ornament frame obtains module, for being carried out using first nerves network handles management image to each image to be managed Processing, obtains at least one dress ornament frame in image to be managed；

Divide module, obtains at least one dress ornament image for dividing at least one dress ornament frame from image to be managed.

Optionally, divide module, be specifically used for carrying out feature extraction to image to be managed, obtain fisrt feature；Based on One feature obtains the alternative frame of at least one of image to be managed；At least one is determined based at least one alternative frame and fisrt feature A dress ornament frame.

Optionally, segmentation module is used when determining at least one dress ornament frame based at least one alternative frame and fisrt feature In handling using pool area layer fisrt feature and at least one alternative frame, at least one dress ornament frame is obtained.

In one or more optional embodiments, attribute obtaining unit 1503 is specifically used for utilizing nervus opticus network Identifying processing is carried out to target dress ornament image, obtains at least one dress ornament attribute of the dress ornament in target dress ornament image.

Identification dress ornament attribute device provided by the embodiments of the present application can use nervus opticus network implementations, the nervus opticus Network, for different attributes, has selected layer different in network to be identified, has improved Attribute Recognition precision when implementing.

Attribute obtaining unit 1503 obtains second feature for handling by shared branch target dress ornament image； Second feature is separately input into a few segmentation branch and at least one classification branch, obtains at least one point of dress ornament image Cut result and at least one classification results；Dress ornament is obtained at least based at least one classification results and at least one segmentation result A kind of dress ornament attribute.

In one or more optional embodiments, device provided by the embodiments of the present application further include:

Quantity statistics unit, for determining the dress ornament for including in image set to be managed based on the quantity of target dress ornament image Quantity.

In one or more optional embodiments, image subset unit 1502 is specifically used for at least two dress ornament figures As being clustered, at least one image subset is obtained；Each image subset determines respectively from least one image subset respectively One dress ornament image obtains at least one target dress ornament image as target dress ornament image.

Wherein, the similarity between every two target dress ornament image is less than setting value.It optionally, can be based at least two clothes Similarity between decorations image clusters at least two dress ornament images, obtains at least one image subset.By similarity compared with Big dress ornament image collection determines the similarity between image two-by-two into an image subset, can be corresponding based on two images The distance between feature (such as: Euclidean distance, COS distance) determine, or by trained neural network directly to extremely Few two dress ornament images are clustered, at least one image subset is obtained.

Optionally, image subset unit 1502 is clustered at least two dress ornament images, obtains at least one image When subset, for being clustered based on the similarity between at least two dress ornament images at least two dress ornament images, obtain extremely A few image subset.

Acquiring unit is matched, is taken at least one dress ornament attribute based on dress ornament and at least one of database of arranging in pairs or groups With combination, acquisition matches dress ornament at least one of clothing matching, and/or obtains and match at least one that dress ornament attribute is arranged in pairs or groups Attribute.

Optionally, acquiring unit is matched, is searched in collocation database at least one dress ornament attribute based on dress ornament It obtains and prestores dress ornament with same or similar dress ornament attribute；Using search obtain prestore the corresponding matching dress ornament of dress ornament as with At least one matching dress ornament of clothing matching.

Optionally, acquiring unit is matched, is searched in collocation database at least one dress ornament attribute based on dress ornament It obtains and prestores dress ornament with same or similar dress ornament attribute；It is obtained based on the corresponding matched combined of dress ornament that prestores obtained is searched At least one matching dress ornament；Using at least one corresponding at least one dress ornament attribute of matching dress ornament of acquisition as with dress ornament attribute At least one match attribute of collocation.

Information acquisition unit, for obtaining dressing relevant information；

Dress ornament recommendation unit, at least one dress ornament attribute based on dressing relevant information and each dress ornament, to pipe It manages screening in image set and obtains at least one recommendation dress ornament or at least a set of recommendation dress ornament combination.

Optionally, dressing relevant information can include but is not limited to following at least one: geographical location information, weather letter Breath, occasion information, chat record information.

Optionally, dress ornament recommendation unit, for obtaining at least one dress ornament attribute, each dressing based on dressing relevant information The corresponding at least one dress ornament attribute of relevant information；Each clothes at least one dress ornament attribute and image set to be managed based on acquisition At least one dress ornament attribute of decorations is matched, and is obtained at least one and is recommended dress ornament.

Optionally, dress ornament recommendation unit each clothes in at least one dress ornament attribute and image set to be managed based on acquisition At least one dress ornament attribute of decorations is matched, when obtaining at least one recommendation dress ornament, in response to the corresponding dress ornament of dress ornament Attribute includes all dress ornament attributes in at least one dress ornament attribute obtained, using dress ornament as recommendation dress ornament.

Optionally, dress ornament recommendation unit, for obtaining at least one dress ornament attribute, each dressing based on dressing relevant information The corresponding at least one dress ornament attribute of relevant information；Each clothes at least one dress ornament attribute and image set to be managed based on acquisition At least one dress ornament attribute of decorations is matched, and is obtained at least two and is recommended dress ornament；At least two based on acquisition recommend dress ornament Obtain at least a set of recommendation dress ornament combination.

Optionally, dress ornament recommendation unit each clothes in at least one dress ornament attribute and image set to be managed based on acquisition At least one dress ornament attribute of decorations is matched, when obtaining at least two recommendation dress ornaments, in response to the corresponding dress ornament of dress ornament Attribute includes all dress ornament attributes in at least one dress ornament attribute obtained, using dress ornament as dress ornament is recommended, obtains at least two A recommendation dress ornament.

Optionally, dress ornament recommendation unit recommends dress ornament to obtain at least a set of recommendation dress ornament group at least two based on acquisition It is to establish matching between at least two recommendation dress ornaments to close for recommending the corresponding dress ornament attribute of dress ornament based at least two when conjunction System；Recommend dress ornament as recommendation dress ornament combination for establish matching relationship at least two.

The other side of the embodiment of the present application additionally provides a kind of electronic equipment, including processor, which includes The dress ornament managing device that above-mentioned any one embodiment provides.

The another aspect of the embodiment of the present application, additionally provides a kind of electronic equipment, comprising: memory, it can for storing It executes instruction；

And processor, for being communicated with the memory to execute the executable instruction to complete above-mentioned any one The operation for the dress ornament management method that embodiment provides.

The still another aspect of the embodiment of the present application additionally provides a kind of computer storage medium, can for storing computer The instruction of reading, the instruction are performed the operation for executing the dress ornament management method that above-mentioned any one embodiment provides.

Another aspect of the embodiment of the present application additionally provides a kind of computer program product, including computer-readable generation Code, when the computer-readable code is run in equipment, the processor in the equipment is executed for realizing above-mentioned any one The instruction for the dress ornament management method that item embodiment provides.

The embodiment of the present application also provides a kind of electronic equipment, such as can be mobile terminal, personal computer (PC), put down Plate computer, server etc..Below with reference to Figure 16, it illustrates the terminal device or the services that are suitable for being used to realize the embodiment of the present application The structural schematic diagram of the electronic equipment 1600 of device: as shown in figure 16, electronic equipment 1600 includes one or more processors, communication Portion etc., one or more of processors for example: one or more central processing unit (CPU) 1601, and/or one or more Application specific processor, application specific processor can be used as accelerator module 1613, it may include but be not limited to image processor (GPU), FPGA, DSP and other asic chip etc application specific processor etc., processor can be according to being stored in read-only memory (ROM)

Executable instruction in 1602 is loaded into random access storage device (RAM) 1603 from storage section 1608 Executable instruction and execute various movements appropriate and processing.Communication unit 1612 may include but be not limited to network interface card, and the network interface card can Including but not limited to IB (Infiniband) network interface card.

Processor can with communicate in read-only memory 1602 and/or random access storage device 1603 to execute executable finger It enables, is connected by bus 1604 with communication unit 1612 and is communicated through communication unit 1612 with other target devices, to complete this Shen Please embodiment provide the corresponding operation of any one method, for example, at least two images to be managed in image set to be managed Dress ornament identification is carried out, at least two dress ornament images are obtained；Each image to be managed includes at least two dress ornament images, each described Dress ornament image corresponds to a dress ornament；At least two dress ornament images are handled, at least one image subset is obtained, based at least One image subset obtains at least one target dress ornament image in image set to be managed；It is obtained based on target dress ornament image each At least one dress ornament attribute of dress ornament in target dress ornament image.

In addition, in RAM 1603, various programs and data needed for being also stored with device operation.CPU1601, ROM1602 and RAM1603 is connected with each other by bus 1604.In the case where there is RAM1603, ROM1602 is optional module.

RAM1603 stores executable instruction, or executable instruction is written into ROM1602 at runtime, executable instruction Central processing unit 1601 is set to execute the corresponding operation of above-mentioned communication means.Input/output (I/O) interface 1605 is also connected to always Line 1604.Communication unit 1612 can integrate setting, may be set to be with multiple submodule (such as multiple IB network interface cards), and In bus link.

I/O interface 1605 is connected to lower component: the importation 1606 including keyboard, mouse etc.；Including such as cathode The output par, c 1607 of ray tube (CRT), liquid crystal display (LCD) etc. and loudspeaker etc.；Storage section including hard disk etc. 1608；

And the communications portion 1609 of the network interface card including LAN card, modem etc..Communications portion 1609 Communication process is executed via the network of such as internet.Driver 1610 is also connected to I/O interface 1605 as needed.Detachably Medium 1611, such as disk, CD, magneto-optic disk, semiconductor memory etc. are mounted on as needed on driver 1610, with Convenient for being mounted into storage section 1608 as needed from the computer program read thereon.

It should be noted that framework as shown in figure 16 is only a kind of optional implementation, it, can root during concrete practice The component count amount and type of above-mentioned Figure 16 are selected, are deleted, increased or replaced according to actual needs；It is set in different function component It sets, separately positioned or integrally disposed and other implementations, such as the separable setting of accelerator module 1613 and CPU1601 can also be used Or accelerator module 1613 can be integrated on CPU1601, the separable setting of communication unit, can also be integrally disposed in CPU1601 or On accelerator module 1613, etc..These interchangeable embodiments each fall within protection scope disclosed in the present application.

Particularly, according to an embodiment of the present application, it may be implemented as computer above with reference to the process of flow chart description Software program.For example, embodiments herein includes a kind of computer program product comprising be tangibly embodied in machine readable Computer program on medium, computer program include the program code for method shown in execution flow chart, program code May include it is corresponding execute the corresponding instruction of method and step provided by the embodiments of the present application, for example, in image set to be managed extremely Few two images to be managed carry out dress ornament identification, obtain at least two dress ornament images；Each image to be managed includes at least two Dress ornament image, each corresponding dress ornament of the dress ornament image；At least two dress ornament images are handled, at least one is obtained Image subset obtains at least one target dress ornament image in image set to be managed based at least one image subset；Based on mesh Mark dress ornament image obtains at least one dress ornament attribute of the dress ornament in each target dress ornament image.In such embodiments, should Computer program can be downloaded and installed from network by communications portion 1609, and/or be pacified from detachable media 1611 Dress.When the computer program is executed by central processing unit (CPU) 1601, the above-mentioned function limited in the present processes is executed The operation of energy.

The present processes and device may be achieved in many ways.For example, can by software, hardware, firmware or Software, hardware, firmware any combination realize the present processes and device.The said sequence of the step of for the method Merely to be illustrated, the step of the present processes, is not limited to sequence described in detail above, special unless otherwise It does not mentionlet alone bright.In addition, in some embodiments, also the application can be embodied as to record program in the recording medium, these programs Including for realizing according to the machine readable instructions of the present processes.Thus, the application also covers storage for executing basis The recording medium of the program of the present processes.

The description of the present application is given for the purpose of illustration and description, and is not exhaustively or by the application It is limited to disclosed form.Many modifications and variations are obvious for the ordinary skill in the art.It selects and retouches Embodiment is stated and be the principle and practical application in order to more preferably illustrate the application, and those skilled in the art is enable to manage Solution the application is to design various embodiments suitable for specific applications with various modifications.

Claims

1. a kind of dress ornament management method characterized by comprising

Dress ornament identification is carried out at least two images to be managed in image set to be managed, obtains at least two dress ornament images；Often A image to be managed includes at least two dress ornament images, each corresponding dress ornament of the dress ornament image；

At least two dress ornaments image is handled, at least one image subset is obtained, based at least one described image Subset obtains at least one target dress ornament image in the image set to be managed；

At least one dress ornament attribute of the dress ornament in each target dress ornament image is obtained based on the target dress ornament image.

2. the method according to claim 1, wherein described to be managed at least two in image set to be managed Image carries out dress ornament identification, obtains at least two dress ornament images, comprising:

To each image to be managed, the image to be managed is handled using first nerves network, obtain it is described to Manage at least one dress ornament frame in image；

3. according to the method described in claim 2, it is characterized in that, described utilize first nerves network to the image to be managed It is handled, obtains at least one dress ornament frame in the image to be managed, comprising:

4. according to the method described in claim 3, it is characterized in that, described based on described at least one alternative frame and described first Feature determines at least one dress ornament frame, comprising:

The fisrt feature and at least one described alternative frame are handled using pool area layer, obtain it is described at least one Dress ornament frame.

5. method according to claim 1 to 4, which is characterized in that described to obtain institute based on the target dress ornament image State at least one dress ornament attribute of the dress ornament in each target dress ornament image, comprising:

Identifying processing is carried out to the target dress ornament image using nervus opticus network, obtains the clothes in the target dress ornament image At least one dress ornament attribute of decorations.

6. a kind of dress ornament managing device characterized by comprising

Dress ornament recognition unit obtains extremely for carrying out dress ornament identification at least two images to be managed in image set to be managed Few two dress ornament images；Each image to be managed includes at least two dress ornament images, each dress ornament image corresponding one A dress ornament；

Image subset unit obtains at least one image subset, is based on for handling at least two dress ornaments image At least one described image subset obtains at least one target dress ornament image in the image set to be managed；

Attribute obtaining unit, for obtaining the dress ornament in each target dress ornament image based on the target dress ornament image extremely A kind of few dress ornament attribute.

7. a kind of electronic equipment, which is characterized in that including processor, the processor includes dress ornament pipe as claimed in claim 6 Manage device.

8. a kind of electronic equipment characterized by comprising memory, for storing executable instruction；

And processor, for being communicated with the memory to execute the executable instruction to complete claim 1 to 5 times The operation for a dress ornament management method of anticipating.

9. a kind of computer storage medium, for storing computer-readable instruction, which is characterized in that described instruction is performed When perform claim require 1 to 5 any one described in dress ornament management method operation.

10. a kind of computer program product, including computer-readable code, which is characterized in that when the computer-readable code When running in equipment, the processor in the equipment is executed for realizing dress ornament management described in claim 1 to 5 any one The instruction of method.