WO2017202086A1 - 图片的筛选方法及装置 - Google Patents

图片的筛选方法及装置 Download PDF

Info

Publication number
WO2017202086A1
WO2017202086A1 PCT/CN2017/074585 CN2017074585W WO2017202086A1 WO 2017202086 A1 WO2017202086 A1 WO 2017202086A1 CN 2017074585 W CN2017074585 W CN 2017074585W WO 2017202086 A1 WO2017202086 A1 WO 2017202086A1
Authority
WO
WIPO (PCT)
Prior art keywords
picture
pictures
specified
similar
feature
Prior art date
Application number
PCT/CN2017/074585
Other languages
English (en)
French (fr)
Inventor
左焘
Original Assignee
中兴通讯股份有限公司
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 中兴通讯股份有限公司 filed Critical 中兴通讯股份有限公司
Priority to EP17801940.2A priority Critical patent/EP3467677A4/en
Publication of WO2017202086A1 publication Critical patent/WO2017202086A1/zh

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/50—Information retrieval; Database structures therefor; File system structures therefor of still image data
    • G06F16/53—Querying
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/50—Information retrieval; Database structures therefor; File system structures therefor of still image data
    • G06F16/51—Indexing; Data structures therefor; Storage structures
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/50—Information retrieval; Database structures therefor; File system structures therefor of still image data
    • G06F16/54—Browsing; Visualisation therefor
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/50—Information retrieval; Database structures therefor; File system structures therefor of still image data
    • G06F16/58—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually
    • G06F16/583—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually using metadata automatically derived from the content

Definitions

  • the present disclosure relates to the field of image processing, and in particular to a method and apparatus for screening a picture.
  • a large number of pictures are stored in a user's terminal (such as a mobile phone), and as the imaging resolution is higher and higher, the picture occupies a large amount of storage space of the terminal.
  • a user's terminal such as a mobile phone
  • the imaging resolution is higher and higher
  • the picture occupies a large amount of storage space of the terminal.
  • users need to browse the gallery to filter and delete.
  • Embodiments of the present disclosure provide a method and apparatus for screening a picture to at least solve the problem that a user spends a lot of time when selecting a similar picture in the related art.
  • a method for filtering a picture including: detecting a trigger event; and acquiring, under the trigger of the trigger event, a picture in the gallery that is similar to a specified picture by a first preset threshold. And a similar picture of the specified picture; taking the specified picture and the similar picture as a set of pictures, and outputting the set of pictures.
  • acquiring a picture in the gallery that is greater than the first preset threshold in the specified image including: acquiring image features of all the pictures in the library; acquiring image features of the other pictures and image features of the specified picture a geometric distance between the other pictures is determined to be a picture whose similarity with the specified picture is greater than a first preset threshold, and the other pictures are used as the Specify a similar image for the image.
  • acquiring image features of all the pictures in the library includes: acquiring semantic features and visual features of all the pictures; and performing feature compression processing on the semantic features and the visual features according to a principal component analysis algorithm to obtain The image features.
  • acquiring the semantic feature and the visual feature of the specified picture including: inputting the specified picture into a deep convolutional neural network model, outputting the semantic feature, and extracting a color of the specified picture, the specified picture
  • the gray level co-occurrence array and the seven invariant moment vectors of the specified picture are used as the visual features.
  • the triggering event includes: a redundant deletion signal for deleting redundant pictures.
  • the method further includes deleting the group of pictures according to a preset rule.
  • One or more pictures are discarded.
  • deleting one or more pictures in the set of pictures according to a preset rule deleting a set of pictures when receiving a delete instruction of deleting a part of the pictures or all pictures in the set of pictures Part or all of the picture.
  • the method further includes: if the redundant deletion signal is not received within a predetermined time, clearing a similar picture list in which the set of pictures is located.
  • a screening apparatus for a picture including: a detecting module, configured to detect a triggering event; and an acquiring module, configured to acquire, in the library and the specified picture, triggered by the triggering event A picture whose similarity is greater than the first preset threshold is used as a similar picture of the specified picture; an output module is configured to use the specified picture and the similar picture as a set of pictures, and output the set of pictures.
  • the acquiring module includes: a first acquiring unit, configured to acquire image features of all the pictures in the library; and a second acquiring unit, configured to acquire image features of the other pictures and image features of the specified picture a geometric distance between the other pictures is determined to be a picture whose similarity with the specified picture is greater than a first preset threshold, and the other pictures are used as the Specify a similar image for the image.
  • the first obtaining unit includes: a first acquiring subunit, configured to acquire semantic features and visual features of all the pictures; and a compressing subunit, configured to perform principal component analysis on the semantic features and the visual features
  • the algorithm performs feature compression processing to obtain the image features.
  • the manner of acquiring the semantic feature and the visual feature of the specified picture comprises: inputting the specified picture into a deep convolutional neural network model, outputting the semantic feature; extracting a color of the specified picture, the designating The gray level co-occurrence array of the picture and the 7 invariant moment vectors of the specified picture are used as the visual features.
  • the triggering event includes: a redundant deletion signal for deleting redundant pictures.
  • the apparatus further includes: a deleting module, configured to delete one or more pictures in the set of pictures according to a preset rule.
  • the deleting module is further configured to delete part or all of the pictures in the group of pictures when receiving a deletion instruction for deleting a part of the pictures or all the pictures in the set of pictures.
  • the similar picture list in which the set of pictures is located is cleared.
  • a storage medium is also provided.
  • the storage medium is configured to store program code for performing the following steps: detecting a trigger event; and acquiring, under the trigger of the trigger event, a picture in the gallery that has a similarity with the specified picture that is greater than a first preset threshold, as the specified a similar picture of the picture; the specified picture and the similar picture are taken as a group of pictures, and the set of pictures is output.
  • a trigger event for example, a signal for deleting a picture
  • a picture whose similarity with the specified picture is greater than the first preset threshold is selected as a similar picture in the gallery, and the specified picture and the similar picture are output.
  • the similar images are stored separately as a whole, which is similar to the classification of photos in the library according to the similarity degree, which solves the problem that the user spends a lot of time when selecting similar pictures, and realizes the rapid screening of similar pictures and the accurate grouping of similar pictures. effect.
  • FIG. 1 is a block diagram showing a hardware structure of a mobile terminal for screening a picture according to an embodiment of the present disclosure
  • FIG. 2 is a flowchart of a method of screening a picture according to an embodiment of the present disclosure
  • FIG. 3 is a structural block diagram 1 of a screening apparatus for a picture according to an embodiment of the present disclosure
  • FIG. 4 is a structural block diagram 2 of a screening apparatus for a picture according to an embodiment of the present disclosure
  • FIG. 5 is a structural block diagram 3 of a screening apparatus for a picture according to an embodiment of the present disclosure
  • FIG. 6 is a structural block diagram 4 of a screening apparatus for a picture according to an embodiment of the present disclosure
  • FIG. 7 is a flowchart of a method for deleting a picture according to an embodiment of the present disclosure.
  • FIG. 8 is a structural diagram of a deletion device of a picture according to an embodiment of the present disclosure.
  • FIG. 9 is a flowchart of a method of deleting a multi-feature based picture according to an embodiment of the present disclosure.
  • FIG. 1 is a hardware structural block diagram of a mobile terminal for screening a picture according to an embodiment of the present disclosure.
  • the mobile terminal 10 may include one or more (only one shown) processor 102 (the processor 102 may include, but is not limited to, a processing device such as a microprocessor MCU or a programmable logic device FPGA).
  • FIG. 1 is merely illustrative and does not limit the structure of the above electronic device.
  • the mobile terminal 10 may also include more or fewer components than those shown in FIG. 1, or have a different configuration than that shown in FIG.
  • the memory 104 can be used to store software programs and modules of the application software, such as program instructions/modules corresponding to the screening method of the pictures in the embodiment of the present disclosure, and the processor 102 executes each by running the software programs and modules stored in the memory 104.
  • a functional application and data processing, that is, the above method is implemented.
  • Memory 104 may include high speed random access memory, and may also include non-volatile memory such as one or more magnetic storage devices, flash memory, or other non-volatile solid state memory.
  • memory 104 may further include memory remotely located relative to processor 102, which may be connected to mobile terminal 10 over a network. Examples of the above networks include but not Limited to the Internet, intranets, local area networks, mobile communication networks, and combinations thereof.
  • the transmission device 106 is configured to transmit a signal that triggers an event.
  • FIG. 2 is a flowchart of a method for screening a picture according to an embodiment of the present disclosure. As shown in FIG. 2, the process includes the following steps. :
  • Step S202 detecting a trigger event
  • the triggering event may be in various forms, for example, detecting that the user is browsing a picture with a specific feature in the terminal, or performing to detect that the user initiates the picture browsing process.
  • the triggering The event may also include a redundant delete signal for deleting redundant pictures, i.e., in the case where a redundant delete signal is detected (i.e., received), screening of the picture is started.
  • Step S204 Acquire, under the triggering of the triggering event, a picture in the gallery that has a similarity with the specified picture that is greater than the first preset threshold, as a similar picture of the specified picture; (subsequent to S206 of FIG. 2)
  • acquiring a picture in the gallery that is similar to the specified picture by the first preset threshold is as follows: acquiring image features of all the pictures in the library; acquiring image features of the other pictures and the designation a geometric distance between the image features of the image, when the geometric distance is less than the second preset threshold, determining that the other image is a picture whose similarity with the specified picture is greater than the first preset threshold, and using the other picture as the Specify a similar image for the image.
  • the above geometric distance is inversely proportional to the similarity, that is, the smaller the geometric distance, the greater the similarity.
  • the obtaining the above similarity may be implemented by a similarity calculation algorithm in the related art, for example:
  • the calculation method of the vector space model is based on the similarity calculation method of the hash method.
  • the similarity can be calculated by: assuming the calculation of the characteristics of any picture, first Obtaining the coordinate points of the picture feature, taking the two-dimensional coordinates as an example, according to the calculation formula of the Euclidean distance Obtaining the Euclidean distance of the feature of the picture and the mixed feature of the specified picture. If the Euclidean distance is less than the second preset threshold, determining that the picture has a high degree of similarity with the specified picture, and outputting the picture as a similar picture of the specified picture.
  • the image features of the picture may be obtained by first acquiring semantic features and visual features.
  • the following deep convolutional neural network models can be used to extract semantic features: the network consists of 5 convolutional layers and 3 fully connected layers.
  • the output of the network is an abstract advanced feature, mainly used for image classification.
  • the training set for deep convolutional neural networks uses the ImgNet data set.
  • the training sample size is 1 million labeled pictures, and the classification category is 1000 categories.
  • the network parameters and network structure used can adopt the parameters and structures in the convolutional neural network model in the related art, and will not be described here. Deep learning combines low-level features to form more abstract high-level representation attribute categories or features to discover distributed feature representations of data.
  • the significant advantage of deep learning is the ability to abstract high-level features and build complex, high-performance models.
  • High-level semantic features are extracted from the input picture using a trained deep convolutional neural network.
  • the deep convolutional neural network model described above can also be applied to the deep convolutional neural network described in the document "ImageNet Classification with Deep Convolutional Neural Networks”.
  • the method of applying BP neural network to natural image classification is used to extract the main color, gray level co-occurrence matrix and 7 invariant moment vectors of the picture as low-level visual features.
  • the extraction method divides the picture into five regions evenly, and extracts the main color, the gray level co-occurrence matrix and the seven invariant moment feature vectors for each small area.
  • Each small area of each picture is extracted with 23-dimensional feature vectors, a total of 5 small areas, that is, a total of 115-dimensional feature vectors are extracted.
  • the low-level features are global and well complemented by high-level features. Combining high-level semantic features and low-level visual features can form a multi-feature structure.
  • the principal component analysis algorithm is used to perform feature compression processing to form mixed features and image features. Calculating the similarity of the picture according to the feature and outputting the similar picture, calculating the geometric distance of the mixed feature of all the pictures in the library and the specified picture, determining that the picture is similar when the geometric distance is less than the second preset threshold, and sequentially outputting each set of pictures.
  • the flow of feature compression processing using a principal component analysis algorithm is described: Suppose we have p maps, and when we extract the semantic features and visual features for each image, we obtain p vectors. The principal component analysis PCA replaces the original n features with a smaller number of m features. The new feature is a linear combination of the old features. These linear combinations maximize the sample variance and try to make the new m features uncorrelated.
  • a 4-step algorithm is used in the PCA process:
  • Step 1 Feature centering. That is, the data for each dimension is subtracted from the mean of the dimension.
  • the "dimension” here refers to a feature (or attribute), and the mean of each dimension becomes 0 after the transformation. Assuming that the original matrix is A, after subtracting the average of the columns for each column, the matrix B is obtained;
  • Step 2 calculating the covariance matrix C of B
  • Step 3 calculating the eigenvalues and eigenvectors of the covariance matrix C;
  • Step 4 Select a feature vector corresponding to the large feature value to obtain a new data set.
  • the deep convolutional neural network is used to generate high-level features, and the image categories are analyzed to ensure the similarity of the output results in the image categories; the low-level features are used to ensure the similarity of the output results on the image content, which is as close as possible to the human senses;
  • Component analysis combines high-level semantic features with low-level visual features, reduces dimensions, reduces redundant features, reduces computational complexity, and meets the need for fast and stable off-line computing.
  • Step S206 of FIG. 2 the specified picture and the similar picture described above are taken as a group of pictures, and the set of pictures is output.
  • the user puts the outputted picture into a similar picture list, receives a redundant deletion signal of the user or another operation signal, and does not receive the user redundancy deletion signal within a predetermined time. Clear the list of similar images for your next use.
  • various applications may be performed for the group of pictures, for example, for browsing, deleting, modifying, etc., in an optional embodiment of the present application, in order to save storage space of the terminal, Delete one or more images in the set of pictures according to a preset rule.
  • the deleting process of the foregoing picture may be implemented in the following manner, but is not limited to: deleting some or all of the pictures in the set of pictures when receiving a deletion instruction for deleting a part of the pictures or all the pictures in the set of pictures. .
  • the trigger event is first detected, and the triggering event triggers the acquisition of the specified image in the gallery.
  • a picture whose similarity is greater than the first preset threshold is used as a similar picture of the specified picture; the specified picture and the similar picture are output as a group of pictures, and the similar picture is processed accurately and timely, which solves a large time for the user to select a similar picture.
  • the problem is to quickly filter out similar images and accurately group similar images.
  • the execution body of the foregoing steps S202-S206 may be a camera, a camera, and a mobile terminal (such as a mobile phone, a tablet) having an image capturing function, but is not limited thereto.
  • the method according to the above embodiment can be implemented by means of software plus a necessary general hardware platform, and of course, by hardware, but in many cases, the former is A better implementation.
  • the technical solution of the present disclosure which is essential or contributes to the related art, may be embodied in the form of a software product stored in a storage medium (such as ROM/RAM, disk, CD-ROM).
  • the instructions include a number of instructions for causing a terminal device (which may be a cell phone, computer, server, or network device, etc.) to perform the methods of various embodiments of the present disclosure.
  • a screening device for a picture is provided, which is used to implement the above-mentioned embodiments and preferred embodiments, and has not been described again.
  • the term "module” may implement a combination of software and/or hardware of a predetermined function.
  • the apparatus described in the following embodiments is preferably implemented in software, hardware, or a combination of software and hardware, is also possible and contemplated.
  • FIG. 3 is a structural block diagram 1 of a screening apparatus for a picture according to an embodiment of the present disclosure. As shown in FIG. 3, the apparatus includes:
  • a detecting module 32 configured to detect a trigger event
  • the obtaining module 34 is connected to the detecting module 32, and is configured to acquire, in the trigger of the triggering event, a picture in the gallery that is similar to the specified picture by the first preset threshold, as a similar picture of the specified picture;
  • the output module 36 is connected to the acquisition module 34 for using the specified picture and the similar picture as a group of pictures, and outputs the group of pictures.
  • FIG. 4 is a structural block diagram 2 of a screening apparatus for a picture according to an embodiment of the present disclosure.
  • the obtaining module 34 includes:
  • a first acquiring unit 42 configured to acquire image features of all the pictures in the gallery
  • the second obtaining unit 44 is connected to the first acquiring unit 42 and configured to acquire a geometric distance between the image feature of the other image and the image feature of the specified image, and determine the other when the geometric distance is less than the second preset threshold.
  • the picture is a picture whose similarity with the specified picture is greater than the first preset threshold, and the other picture is used as a similar picture of the specified picture.
  • FIG. 5 is a structural block diagram 3 of a screening apparatus for a picture according to an embodiment of the present disclosure.
  • the first obtaining unit 42 includes:
  • a first obtaining subunit 52 configured to acquire semantic features and visual features of the all pictures
  • the compression sub-unit 54 is coupled to the first acquisition sub-unit 52 for performing feature compression processing on the semantic feature and the visual feature according to a principal component analysis algorithm to obtain the image feature.
  • the manner of obtaining the semantic feature and the visual feature of the specified picture comprises: inputting the specified picture into a deep convolutional neural network model, and outputting the semantic feature; extracting a color of the specified picture, and gray level symbiosis of the specified picture.
  • the array and the seven invariant moment vectors of the specified picture are used as the visual features.
  • the trigger event includes: a redundancy delete signal for deleting redundant pictures.
  • FIG. 6 is a structural block diagram 4 of a screening apparatus for a picture according to an embodiment of the present disclosure. As shown in FIG. 6, the apparatus includes:
  • the deleting module 62 is connected to the output module 36, and after outputting the set of pictures, delete one or more pictures in the set of pictures according to a preset rule.
  • the deleting module 62 is further configured to delete part or all of the pictures in the set of pictures when receiving a deletion instruction for deleting a part of the pictures or all the pictures in the set of pictures.
  • the similar picture list in which the set of pictures is located is cleared.
  • each of the above modules may be implemented by software or hardware.
  • the foregoing may be implemented by, but not limited to, the foregoing modules are all located in the same processor; or, the above modules are in any combination.
  • the forms are located in different processors.
  • FIG. 7 is a flowchart of a method for deleting a picture according to an embodiment of the present disclosure. As shown in FIG. 7, the steps are as follows:
  • Step S701 (corresponding to the function of the detecting module 32 in the second embodiment), the user starts a similar picture deletion process, and sends a redundancy delete signal;
  • Step S702 (corresponding to the function of the obtaining module 34 in the above embodiment 2) analyzing the library with a similar feature deletion method based on multiple features, and obtaining a similar photo list;
  • Step S703 (corresponding to the function of the output module 36 in the above embodiment 2), is displayed in groups as a group, and displayed in groups;
  • Step S704 (corresponding to the function of the deleting module 62 in the second embodiment), the user selects the picture to be deleted, and deletes the picture from the library; if the user does not perform the deleting operation, the input operation is cancelled, at this time, Clear the list of similar images for your next use.
  • FIG. 8 is a structural diagram of a device for deleting a picture according to an embodiment of the present disclosure.
  • the device is used in a storage space cleaning system of a digital device to search for similar pictures in a gallery to improve the cleaning speed of the user. , reduce the use of storage space.
  • the apparatus includes a gallery 802, a similar matching module 804 coupled to the gallery 802, a similar list processing module 806 coupled to the similar matching module 804, and a deletion response module 808 coupled to the similar list processing module 806.
  • Gallery 802 (corresponding to the memory 104 in the above embodiment 1) for receiving and storing a user picture
  • the similar matching module 804 (corresponding to the functions of the detecting module 32 and the obtaining module 34 in the foregoing embodiment 2), is configured to receive a redundant deletion signal of the user, and use a similar feature deletion method based on multiple features to divide from the gallery 802. Parsing a similar picture, sent to the similar list processing module 806;
  • the similar list processing module 806, (corresponding to the function of the output module 36 in the above embodiment 2), is for receiving similar pictures, and outputting pictures in groups, and the pictures of each group are similar;
  • the deletion response module 808, (corresponding to the function of the deletion module 62 in the above embodiment 2), after the user selects a similar picture to be deleted, sends a picture deletion signal, and the deletion module 808 receives the signal and deletes the corresponding picture from the gallery. .
  • FIG. 9 is a flowchart of a method for deleting a multi-feature based picture according to an embodiment of the present disclosure. As shown in FIG. 9, the steps are as follows:
  • Step S901 high-level semantic feature extraction, using the deep convolutional neural network described in the document "ImageNet Classification with Deep Convolutional Neural Networks”.
  • the network consists of 5 convolutional layers and 3 fully connected layers.
  • the output of the network is an abstract high-level feature, mainly used for image classification.
  • the training set for deep convolutional neural networks uses the ImgNet data set.
  • the training sample size is 1 million labeled images, and the classification category is 1000 categories.
  • the network parameters and network structure used are the same as the papers.
  • Step S902 low-level visual feature extraction, using the method in the literature "Applying BP neural network to natural image classification", extracting the main color of the image, the gray level co-occurrence matrix and the seven invariant moment vectors as low-level visual features.
  • the extraction method divides the image into five regions uniformly, and extracts the main color, the gray level co-occurrence matrix and the seven invariant moment feature vectors for each small region.
  • a 23-dimensional feature vector is extracted for each small area of each image, for a total of 5 small regions, so a total of 115-dimensional feature vectors are extracted.
  • Step S903 using a principal component analysis algorithm to perform feature compression on the extracted high-level semantic features and low-level visual features to form a hybrid feature.
  • the goal is to reduce feature dimensions and reduce feature redundancy.
  • Step S904 calculating image similarity according to the feature and outputting the similar image, calculating the geometric distance of the mixed features of all the images in the library and each of the other images, and determining that the geometric distance is smaller than the threshold, and each group of similar pictures is similar. Output in order.
  • Embodiments of the present disclosure also provide a storage medium.
  • the foregoing storage medium may be configured to store program code for performing the following steps:
  • the specified picture and the similar picture are taken as a group of pictures, and the group of pictures is output.
  • the foregoing storage medium may include, but not limited to, a USB flash drive, a Read-Only Memory (ROM), a Random Access Memory (RAM), a mobile hard disk, and a magnetic memory.
  • ROM Read-Only Memory
  • RAM Random Access Memory
  • a mobile hard disk e.g., a hard disk
  • magnetic memory e.g., a hard disk
  • the processor executes the foregoing embodiment according to the stored program code in the storage medium. Method steps carried.
  • modules or steps of the present disclosure described above can be implemented by a general-purpose computing device that can be centralized on a single computing device or distributed across a network of multiple computing devices. Alternatively, they may be implemented by program code executable by the computing device such that they may be stored in the storage device by the computing device and, in some cases, may be different from the order herein.
  • the steps shown or described are performed, or they are separately fabricated into individual integrated circuit modules, or a plurality of modules or steps thereof are fabricated as a single integrated circuit module. As such, the disclosure is not limited to any specific combination of hardware and software.
  • the present disclosure is applicable to the field of image processing, and solves the problem that a user spends a lot of time when selecting a similar picture, and realizes the effect of quickly filtering out similar pictures and accurately grouping similar pictures.

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Data Mining & Analysis (AREA)
  • Databases & Information Systems (AREA)
  • Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Library & Information Science (AREA)
  • Software Systems (AREA)
  • Image Analysis (AREA)
  • Information Retrieval, Db Structures And Fs Structures Therefor (AREA)

Abstract

一种图片的筛选方法及装置，其中，该方法包括：检测触发事件（S202），在该触发事件触发下，获取图库中与指定图片的相似度大于第一预设阈值的图片，作为该指定图片的相似图片（S204）；将该指定图片和该相似图片作为一组图片输出（S206）。采用上述方法和装置可以准确及时处理相似图片，解决了用户选择相似图片时耗费大量时间的问题，实现了快速筛选出相似图片，以及将相似图片准确分组的效果。

Description

图片的筛选方法及装置 技术领域
本公开涉及图像处理领域，具体而言，涉及一种图片的筛选方法及装置。
背景技术
在相关技术中，用户的终端(如手机等)中都存有大量的图片，随着摄像分辨率越来越高，图片占据了终端大量的存储空间。在用户拍照过程中，常有一些相似的图片产生，比如连拍、重拍。为了留下质量较好的图片，节省存储空间，用户需要通过浏览图库进行筛选，进而删除。
如果用户清理图库中的相似图片，需要一张张浏览整个图库，花费大量的时间和精力。如果不清理，相似图片会占据大量存储空间，用户浏览图库时，筛选相似图片会花费用户更多的操作以掠过或删除。
针对相关技术中，用户选择相似图片时耗费大量时间的问题，目前还没有有效的解决方案。
发明内容
本公开实施例提供了一种图片的筛选方法及装置，以至少解决相关技术中用户选择相似图片时耗费大量时间的问题。
根据本公开的一个实施例，提供了一种图片的筛选方法，包括：检测触发事件；在所述触发事件的触发下，获取图库中与指定图片的相似度大于第一预设阈值的图片，作为所述指定图片的相似图片；将所述指定图片和所述相似图片作为一组图片，并输出该组图片。
可选地，获取图库中与指定图片的相似度大于第一预设阈值的图片，包括：获取所述图库中所有图片的图像特征；获取其它图片的图像特征与所述指定图片的图像特征之间的几何距离，在所述几何距离小于第二预设阈值时，确定所述其它图片为与所述指定图片的相似度大于第一预设阈值的图片，并将所述其它图片作为所述指定图片的相似图片。
可选地，获取所述图库中所有图片的图像特征，包括：获取所述所有图片的语义特征和视觉特征；对所述语义特征和所述视觉特征按照主成分分析算法进行特征压缩处理，得到所述图像特征。
可选地，获取所述指定图片的语义特征和视觉特征，包括：将所述指定图片输入深度卷积神经网络模型，输出得到所述语义特征；提取所述指定图片的颜色、所述指定图片的灰度共生阵以及所述指定图片的7个不变矩向量作为所述视觉特征。
可选地，所述触发事件包括：用于删除冗余图片的冗余删除信号。
可选地，输出该组图片之后，所述方法还包括：按照预设规则删除所述一组图片中的 一个或多个图片。
可选地，按照预设规则删除所述一组图片中的一个或多个图片：在接收到删除所述一组图片中的部分图片或全部图片的删除指令时，删除所述一组图片中的部分或全部图片。
可选地，所述方法还包括：在预定时间内没有接收到所述冗余删除信号的情况下，清空所述一组图片所在的相似图片列表。
根据本公开的另一个实施例，提供了一种图片的筛选装置，包括：检测模块，用于检测触发事件；获取模块，用于在所述触发事件的触发下，获取图库中与指定图片的相似度大于第一预设阈值的图片，作为所述指定图片的相似图片；输出模块，用于将所述指定图片和所述相似图片作为一组图片，并输出该组图片。
可选地，所述获取模块包括：第一获取单元，用于获取所述图库中所有图片的图像特征；第二获取单元，用于获取其它图片的图像特征与所述指定图片的图像特征之间的几何距离，在所述几何距离小于第二预设阈值时，确定所述其它图片为与所述指定图片的相似度大于第一预设阈值的图片，并将所述其它图片作为所述指定图片的相似图片。
可选地，第一获取单元包括：第一获取子单元，用于获取所述所有图片的语义特征和视觉特征；压缩子单元，用于对所述语义特征和所述视觉特征按照主成分分析算法进行特征压缩处理，得到所述图像特征。
可选地，获取所述指定图片的语义特征和视觉特征的方式包括：将所述指定图片输入深度卷积神经网络模型，输出得到所述语义特征；提取所述指定图片的颜色、所述指定图片的灰度共生阵以及所述指定图片的7个不变矩向量作为所述视觉特征。
可选地，所述触发事件包括：用于删除冗余图片的冗余删除信号。
可选地，输出该组图片之后，所述装置还包括：删除模块，用于按照预设规则删除所述一组图片中的一个或多个图片。
可选地，所述删除模块还用于在接收到删除所述一组图片中的部分图片或全部图片的删除指令时，删除所述一组图片中的部分或全部图片。
可选地，在预定时间内没有接收到所述冗余删除信号的情况下，清空所述一组图片所在的相似图片列表。
根据本公开的又一个实施例，还提供了一种存储介质。该存储介质设置为存储用于执行以下步骤的程序代码：检测触发事件；在所述触发事件的触发下，获取图库中与指定图片的相似度大于第一预设阈值的图片，作为所述指定图片的相似图片；将所述指定图片和所述相似图片作为一组图片，并输出该组图片。
通过本公开，在检测到触发事件(例如用于删除图片的信号)时，在图库中选取与指定图片的相似度大于第一预设阈值的图片作为相似图片，并输出该指定图片和相似图片，即将相似图片作为一个整体单独存储，类似于将图库中相片依据相似度进行分类，解决了用户选择相似图片时耗费大量时间的问题，实现了快速筛选出相似图片，以及将相似图片准确分组的效果。
附图说明
此处所说明的附图用来提供对本公开的进一步理解，构成本申请的一部分，本公开的示意性实施例及其说明用于解释本公开，并不构成对本公开的不当限定。在附图中：
图1是本公开实施例的一种图片的筛选方法的移动终端的硬件结构框图；
图2是根据本公开实施例的一种图片的筛选方法的流程图；
图3是根据本公开实施例的一种图片的筛选装置的结构框图一；
图4是根据本公开实施例的一种图片的筛选装置的结构框图二；
图5是根据本公开实施例的一种图片的筛选装置的结构框图三；
图6是根据本公开实施例的一种图片的筛选装置的结构框图四；
图7是根据本公开实施例的图片的删除方法流程图；
图8是根据本公开实施例的图片的删除装置的结构图；
图9是根据本公开实施例的一种基于多特征的图片的删除方法的流程图。
具体实施方式
下文中将参考附图并结合实施例来详细说明本公开。需要说明的是，在不冲突的情况下，本申请中的实施例及实施例中的特征可以相互组合。
需要说明的是，本公开的说明书和权利要求书及上述附图中的术语“第一”、“第二”等是用于区别类似的对象，而不必用于描述特定的顺序或先后次序。
实施例1
本申请实施例1所提供的方法实施例可以在移动终端、相机、计算机终端或者类似的运算装置中执行。以运行在移动终端上为例，图1是本公开实施例的一种图片的筛选方法的移动终端的硬件结构框图。如图1所示，移动终端10可以包括一个或多个(图中仅示出一个)处理器102(处理器102可以包括但不限于微处理器MCU或可编程逻辑器件FPGA等的处理装置)、用于存储数据的存储器104、以及用于通信功能的传输装置106。本领域普通技术人员可以理解，图1所示的结构仅为示意，其并不对上述电子装置的结构造成限定。例如，移动终端10还可包括比图1中所示更多或者更少的组件，或者具有与图1所示不同的配置。
存储器104可用于存储应用软件的软件程序以及模块，如本公开实施例中的图片的筛选方法对应的程序指令/模块，处理器102通过运行存储在存储器104内的软件程序以及模块，从而执行各种功能应用以及数据处理，即实现上述的方法。存储器104可包括高速随机存储器，还可包括非易失性存储器，如一个或者多个磁性存储装置、闪存、或者其他非易失性固态存储器。在一些实例中，存储器104可进一步包括相对于处理器102远程设置的存储器，这些远程存储器可以通过网络连接至移动终端10。上述网络的实例包括但不 限于互联网、企业内部网、局域网、移动通信网及其组合。
传输装置106用于传输触发事件的信号。
在本实施例中提供了一种运行于上述移动终端的图片的筛选方法，图2是根据本公开实施例的一种图片的筛选方法的流程图，如图2所示，该流程包括如下步骤：
步骤S202，检测触发事件；
该触发事件可以表现为多种形式，例如检测到用户正在浏览终端中具有特定特征的图片，也可以表现为检测到用户启动了图片浏览进程，在本公开的一个可选实施例中，上述触发事件还可以包括用于删除冗余图片的冗余删除信号，即在检测到(即接收到)冗余删除信号的情况下，开始进行图片的筛选。
步骤S204，在上述触发事件的触发下，获取图库中与指定图片的相似度大于第一预设阈值的图片，作为该指定图片的相似图片；(后续有图2的S206)
在本公开的一个可选实施例中，获取图库中与指定图片的相似度大于第一预设阈值的图片方式如下：获取该图库中所有图片的图像特征；获取其它图片的图像特征与该指定图片的图像特征之间的几何距离，在该几何距离小于第二预设阈值时，确定该其它图片为与该指定图片的相似度大于第一预设阈值的图片，并将该其它图片作为该指定图片的相似图片。由此也可以看出，上述几何距离是与相似度成反比的，即几何距离越小，上述相似度越大。其中，对于上述相似度的获取可以通过相关技术中的相似度计算算法实现，例如：
向量空间模型的计算方法，基于hash方法的相似计算方法，以空间向量模型的计算方法中的欧式距离计算方式为例，该相似度的计算方式可以是：假设计算任一张图片的特征，首先，获取所述图片特征的坐标点，以二维坐标为例，按照欧式距离的计算公式
Figure PCTCN2017074585-appb-000001
得到图片的特征与该指定图片的混合特征的欧式距离，如果欧式距离小于第二预设阈值时，则判断该张图片与指定图片的相似度高，输出该张图片作为指定图片的相似图片。
在本申请的一个可选实施例中，可以通过以下方式获取图片的图像特征：首先获取语义特征和视觉特征。提取语义特征时可以采用以下深度卷积神经网络模型实现：网络由5个卷积层和3个全连接层构成，网络的输出是抽象的高级特征，主要用于图片分类。深度卷积神经网络的训练集采用ImgNet数据集。训练样本量为100万张有标注图片，分类类别为1000个类别，所用的网络参数和网络结构可以采用相关技术中的卷积神经网络模型中的参数和结构，此处不再赘述。深度学习通过组合低层特征形成更加抽象的高层表示属性类别或特征，以发现数据的分布式特征表示。深度学习显著的优点是可以抽象出高级特征，构建出复杂高性能的模型。使用已训练的深度卷积神经网络，对输入图片提取高层语义特征。上述深度卷积神经网络模型也可以采用文献《ImageNet Classification with Deep Convolutional Neural Networks》中记载的深度卷积神经网络。
并且，提取视觉特征时采用文献《应用BP神经网络对自然图像分类》中的方法，提取图片的主要颜色、灰度共生矩阵和7个不变矩向量作为低层视觉特征。该提取方法对图片均匀分割成5个区域，再对每个小区域分别提取主要颜色、灰度共生矩阵和7个不变矩的特征向量。每幅图片的每个小区域提取了23维特征向量，共5个小区域，即共提取了115维特征向量。低层特征具有全局性，很好地作为高层特征的补充。联合高层语义特征和低层视觉特征可以形成一种多特征结构。
在获取了语义特征和视觉特征之后，使用主成分分析算法进行特征压缩处理，形成混合特征及图像特征。根据特征计算图片相似度并输出相似图片，计算图库中所有图片和指定图片的混合特征的几何距离，在该几何距离小于第二预设阈值时确定图片相似，将每一组图片依次输出。在上述实施例的一个可选实施例中记载了使用主成分分析算法进行特征压缩处理的流程：假设我们有p张图，当我们对每一张图片完成语义特征和视觉特征的提取之后，得到p个向量。主成分分析PCA把原先的n个特征用数目更少的m个特征取代，新特征是旧特征的线性组合，这些线性组合最大化样本方差，尽量使新的m个特征互不相关。在PCA过程中采用了4步算法：
步骤1，特征中心化。即每一维的数据都减去该维的均值。这里的“维”指的就是一个特征(或属性)，变换之后每一维的均值都变成了0。假设原始矩阵是A，每一列减去该列均值后，得到矩阵B；
步骤2，计算B的协方差矩阵C；
步骤3，计算协方差矩阵C的特征值和特征向量；
步骤4，选取大的特征值对应的特征向量，得到新的数据集。
完成以上一系列过程之后，我们就得到p个降维后的向量。
在上述实施例中利用深度卷积神经网络产生高层特征，对图片类别分析，保证输出结果在图片类别上的相似；利用低层特征保证输出结果在图片内容上的相似，尽可能符合人类感官；主成分分析，将高层语义特征和低层视觉特征融合，降低维度，减少冗余特征，减轻计算量，满足离线计算的快速稳定的需求。
图2的步骤S206，将上述指定图片和上述相似图片作为一组图片，并输出该组图片。
在本公开的一个可选实施例中，用户将输出后的图片放入相似图片列表，接收用户的冗余删除信号或者别的操作信号，在预定时间没有接收到用户冗余删除信号的情况下，清空该相似图片列表，以备下次使用。
在通过步骤S206输出该组图片之后，对于该组图片可以进行各种应用，例如可以用于浏览，删除，修改等，在本申请一个可选实施例中，为了节省终端的存储空间，还可以按照预设规则删除该一组图片中的一个或多个图片。具体地，可以通过以下方式实现上述图片的删除过程，但不限于此：在接收到删除该一组图片中的部分图片或全部图片的删除指令时，删除该一组图片中的部分或全部图片。
通过上述步骤，首先检测触发事件，在该触发事件触发下，获取图库中与指定图片的 相似度大于第一预设阈值的图片，作为该指定图片的相似图片；将该指定图片和该相似图片作为一组图片输出，准确及时处理了相似图片，解决了用户选择相似图片时耗费大量时间的问题，实现了快速筛选出相似图片，以及将相似图片准确分组的效果。
可选地，上述步骤S202-S206的执行主体可以为照相机、摄像机，以及具有图像采集功能的移动终端(例如手机、平板电脑)等，但不限于此。
通过以上的实施方式的描述，本领域的技术人员可以清楚地了解到根据上述实施例的方法可借助软件加必需的通用硬件平台的方式来实现，当然也可以通过硬件，但很多情况下前者是更佳的实施方式。基于这样的理解，本公开的技术方案本质上或者说对相关技术做出贡献的部分可以以软件产品的形式体现出来，该计算机软件产品存储在一个存储介质(如ROM/RAM、磁碟、光盘)中，包括若干指令用以使得一台终端设备(可以是手机，计算机，服务器，或者网络设备等)执行本公开各个实施例的方法。
实施例2
在本实施例中还提供了一种图片的筛选装置，该装置用于实现上述实施例及优选实施方式，已经进行过说明的不再赘述。如以下所使用的，术语“模块”可以实现预定功能的软件和/或硬件的组合。尽管以下实施例所描述的装置较佳地以软件来实现，但是硬件，或者软件和硬件的组合的实现也是可能并被构想的。
图3是根据本公开实施例的一种图片的筛选装置的结构框图一，如图3所示，该装置包括：
检测模块32，用于检测触发事件；
获取模块34，与检测模块32连接，用于在该触发事件的触发下，获取图库中与指定图片的相似度大于第一预设阈值的图片，作为该指定图片的相似图片；
输出模块36，与获取模块34连接，用于将该指定图片和该相似图片作为一组图片，并输出该组图片。
图4是根据本公开实施例的一种图片的筛选装置的结构框图二，如图4所示，该获取模块34包括：
第一获取单元42，用于获取该图库中所有图片的图像特征；
第二获取单元44，与第一获取单元42连接，用于获取其它图片的图像特征与该指定图片的图像特征之间的几何距离，在该几何距离小于第二预设阈值时，确定该其它图片为与该指定图片的相似度大于第一预设阈值的图片，并将该其它图片作为该指定图片的相似图片。
图5是根据本公开实施例的一种图片的筛选装置的结构框图三，如图5所示，第一获取单元42包括：
第一获取子单元52，用于获取该所有图片的语义特征和视觉特征；
压缩子单元54，与第一获取子单元52连接，用于对该语义特征和该视觉特征按照主成分分析算法进行特征压缩处理，得到该图像特征。
可选地，获取该指定图片的语义特征和视觉特征的方式包括：将该指定图片输入深度卷积神经网络模型，输出得到该语义特征；提取该指定图片的颜色、该指定图片的灰度共生阵以及该指定图片的7个不变矩向量作为该视觉特征。
可选地，该触发事件包括：用于删除冗余图片的冗余删除信号。
图6是根据本公开实施例的一种图片的筛选装置的结构框图四，如图6所示，该装置包括：
删除模块62，与输出模块36连接，用于输出该组图片之后，按照预设规则删除该一组图片中的一个或多个图片。
可选地，该删除模块62还用于在接收到删除该一组图片中的部分图片或全部图片的删除指令时，删除该一组图片中的部分或全部图片。
可选地，在预定时间内没有接收到该冗余删除信号的情况下，清空该一组图片所在的相似图片列表。
需要说明的是，上述各个模块是可以通过软件或硬件来实现的，对于后者，可以通过以下方式实现，但不限于此：上述模块均位于同一处理器中；或者，上述各个模块以任意组合的形式分别位于不同的处理器中。
实施例3
本公开的实施还提供了一种图片的删除装置，图7是根据本公开实施例的图片的删除方法流程图，如图7所示，步骤如下：
步骤S701，(相当于上述实施例2中的检测模块32的功能)用户开启相似图片删除流程，发送冗余删除信号；
步骤S702，(相当于上述实施例2中的获取模块34的功能)以基于多特征的相似图片删除方法分析图库，获取相似相片列表；
步骤S703，(相当于上述实施例2中的输出模块36的功能)以互相相似的图片为一组，按组输出显示；
步骤S704，(相当于上述实施例2中的删除模块62的功能)用户选定要删除的图片，从图库中删除；若用户没有执行删除操作，而是取消了本次输入操作，此时，清空相似图片列表，以备下次使用。
图8是根据本公开实施例的图片的删除装置的结构图，如图8所示，该装置用于数码设备的存储空间清理系统当中，用于查找图库中的相似图片，提高用户的清理速度，降低存储空间的使用率。该装置包括：图库802，与图库802连接的相似匹配模块804，与相似匹配模块804连接的相似列表处理模块806，与相似列表处理模块806连接的删除响应模块808。
图库802，(相当于上述实施例1中的存储器104)用于接收并存储用户图片；
相似匹配模块804，(相当于上述实施例2中的检测模块32和获取模块34的功能)用于接收用户的冗余删除信号，并使用基于多特征的相似图片删除方法，从图库802中分 析出相似的图片，发送给相似列表处理模块806；
相似列表处理模块806，(相当于上述实施例2中的输出模块36的功能)用于接收相似图片，并以组为单位输出图片，每一组的图片之间是相似的；
删除响应模块808，(相当于上述实施例2中的删除模块62的功能)用户选定需要删除的相似图片之后，发送图片删除信号，删除模块808接收该信号，并从图库中删除相应的图片。
实施例4
图9是根据本公开实施例的一种基于多特征的图片的删除方法的流程图，如图9所示，步骤如下：
步骤S901，高层语义特征提取，采用文献《ImageNet Classification with Deep Convolutional Neural Networks》中所述的深度卷积神经网络。网络由5个卷积层和3个全连接层构成，网络的输出是抽象的高级特征，主要用于图像分类。深度卷积神经网络的训练集采用ImgNet数据集。训练样本量为100万张有标注图像，分类类别为1000个类别，所用的网络参数和网络结构和论文相同。
步骤S902，低层视觉特征提取，采用文献《应用BP神经网络对自然图像分类》中的方法，提取图像的主要颜色、灰度共生矩阵和7个不变矩向量作为低层视觉特征。该提取方法对图像均匀分割成5个区域，再对每个小区域分别提取主要颜色、灰度共生矩阵和7个不变矩的特征向量。每副图像的每个小区域提取了23维特征向量，共5个小区域，所以一共提取了115维特征向量。
步骤S903，采用主成分分析算法，对提取的高层语义特征和低层视觉特征进行特征压缩，形成混合特征。目的是降低特征维度，减少特征冗余。
步骤S904，根据特征计算图像相似度并输出相似图像，计算图库中所有图像和其他每一幅图像的混合特征的几何距离，判定几何距离小于阈值的图片之间为相似，将每一组相似图片依次输出。
实施例5
本公开的实施例还提供了一种存储介质。可选地，在本实施例中，上述存储介质可以被设置为存储用于执行以下步骤的程序代码：
S1，检测触发事件；
S2，在该触发事件的触发下，获取图库中与指定图片的相似度大于第一预设阈值的图片，作为该指定图片的相似图片；
S3，将该指定图片和该相似图片作为一组图片，并输出该组图片。
可选地，在本实施例中，上述存储介质可以包括但不限于：U盘、只读存储器(ROM，Read-Only Memory)、随机存取存储器(RAM，Random Access Memory)、移动硬盘、磁碟或者光盘等各种可以存储程序代码的介质。
可选地，在本实施例中，处理器根据存储介质中已存储的程序代码执行上述实施例记 载的方法步骤。
可选地，本实施例中的具体示例可以参考上述实施例及可选实施方式中所描述的示例，本实施例在此不再赘述。
显然，本领域的技术人员应该明白，上述的本公开的各模块或各步骤可以用通用的计算装置来实现，它们可以集中在单个的计算装置上，或者分布在多个计算装置所组成的网络上，可选地，它们可以用计算装置可执行的程序代码来实现，从而，可以将它们存储在存储装置中由计算装置来执行，并且在某些情况下，可以以不同于此处的顺序执行所示出或描述的步骤，或者将它们分别制作成各个集成电路模块，或者将它们中的多个模块或步骤制作成单个集成电路模块来实现。这样，本公开不限制于任何特定的硬件和软件结合。
以上所述仅为本公开的优选实施例而已，并不用于限制本公开，对于本领域的技术人员来说，本公开可以有各种更改和变化。凡在本公开的精神和原则之内，所作的任何修改、等同替换、改进等，均应包含在本公开的保护范围之内。
工业实用性
本公开适用于图像处理领域，用以解决用户选择相似图片时耗费大量时间的问题，实现快速筛选出相似图片，以及将相似图片准确分组的效果。

Claims (14)

  1. 一种图片的筛选方法，包括：
    检测触发事件；
    在所述触发事件的触发下，获取图库中与指定图片的相似度大于第一预设阈值的图片，作为所述指定图片的相似图片；
    将所述指定图片和所述相似图片作为一组图片，并输出该组图片。
  2. 根据权利要求1所述的方法，其中获取图库中与指定图片的相似度大于第一预设阈值的图片，包括：
    获取所述图库中所有图片的图像特征；
    获取其它图片的图像特征与所述指定图片的图像特征之间的几何距离，在所述几何距离小于第二预设阈值时，确定所述其它图片为与所述指定图片的相似度大于第一预设阈值的图片，并将所述其它图片作为所述指定图片的相似图片。
  3. 根据权利要求2所述的方法，其中获取所述图库中所有图片的图像特征，包括：
    获取所述所有图片的语义特征和视觉特征；
    对所述语义特征和所述视觉特征按照主成分分析算法进行特征压缩处理，得到所述图像特征。
  4. 根据权利要求3所述的方法，其中获取所述指定图片的语义特征和视觉特征，包括：
    将所述指定图片输入深度卷积神经网络模型，输出得到所述语义特征；
    提取所述指定图片的颜色、所述指定图片的灰度共生阵以及所述指定图片的7个不变矩向量作为所述视觉特征。
  5. 根据权利要求1至4中任一项所述的方法，其中所述触发事件包括：用于删除冗余图片的冗余删除信号。
  6. 根据权利要求5所述的方法，其中输出该组图片之后，所述方法还包括：按照预设规则删除所述一组图片中的一个或多个图片。
  7. 根据权利要求6所述的方法，其中按照预设规则删除所述一组图片中的一个或多个图片包括：
    在接收到删除所述一组图片中的部分图片或全部图片的删除指令时，删除所述一组图片中的部分或全部图片。
  8. 根据权利要求5所述的方法，其中所述方法还包括：在预定时间内没有接收到所述冗余删除信号的情况下，清空所述一组图片所在的相似图片列表。
  9. 一种图片的筛选装置，包括：
    检测模块，设置为检测触发事件；
    获取模块，设置为在所述触发事件的触发下，获取图库中与指定图片的相似度大于第一预设阈值的图片，作为所述指定图片的相似图片；
    输出模块，设置为将所述指定图片和所述相似图片作为一组图片，并输出该组图片。
  10. 根据权利要求9所述的装置，其中所述获取模块包括：
    第一获取单元，设置为获取所述图库中所有图片的图像特征；
    第二获取单元，设置为获取其它图片的图像特征与所述指定图片的图像特征之间的几何距离，在所述几何距离小于第二预设阈值时，确定所述其它图片为与所述指定图片的相似度大于第一预设阈值的图片，并将所述其它图片作为所述指定图片的相似图片。
  11. 根据权利要求10所述的装置，其中第一获取单元包括：
    第一获取子单元，设置为获取所述所有图片的语义特征和视觉特征；
    压缩子单元，设置为对所述语义特征和所述视觉特征按照主成分分析算法进行特征压缩处理，得到所述图像特征。
  12. 根据权利要求11所述的装置，其中获取所述指定图片的语义特征和视觉特征的方式包括：
    将所述指定图片输入深度卷积神经网络模型，输出得到所述语义特征；
    提取所述指定图片的颜色、所述指定图片的灰度共生阵以及所述指定图片的7个不变矩向量作为所述视觉特征。
  13. 根据权利要求9至12中任一项所述的装置，其中所述触发事件包括：设置为删除冗余图片的冗余删除信号。
  14. 根据权利要求13所述的装置，其中输出该组图片之后，所述装置还包括：
    删除模块，设置为按照预设规则删除所述一组图片中的一个或多个图片。
PCT/CN2017/074585 2016-05-23 2017-02-23 图片的筛选方法及装置 WO2017202086A1 (zh)

Priority Applications (1)

Application Number Priority Date Filing Date Title
EP17801940.2A EP3467677A4 (en) 2016-05-23 2017-02-23 IMAGE SCREENING METHOD AND DEVICE

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201610346355.5 2016-05-23
CN201610346355.5A CN107423297A (zh) 2016-05-23 2016-05-23 图片的筛选方法及装置

Publications (1)

Publication Number Publication Date
WO2017202086A1 true WO2017202086A1 (zh) 2017-11-30

Family

ID=60411054

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2017/074585 WO2017202086A1 (zh) 2016-05-23 2017-02-23 图片的筛选方法及装置

Country Status (3)

Country Link
EP (1) EP3467677A4 (zh)
CN (1) CN107423297A (zh)
WO (1) WO2017202086A1 (zh)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111274421A (zh) * 2020-01-15 2020-06-12 平安科技（深圳）有限公司 图片数据清洗方法、装置、计算机设备和存储介质

Families Citing this family (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108228844B (zh) * 2018-01-09 2020-10-27 美的集团股份有限公司 一种图片筛选方法及装置、存储介质、计算机设备
CN110727815B (zh) * 2018-07-16 2024-01-30 浙江大华技术股份有限公司 一种更新以图搜图的图库的方法、设备及可读存储介质
CN112926487B (zh) * 2021-03-17 2022-02-11 西安电子科技大学广州研究院 一种行人重识别的方法和装置

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20050105803A1 (en) * 2003-11-19 2005-05-19 Ray Lawrence A. Method for selecting an emphasis image from an image collection based upon content recognition
CN103558996A (zh) * 2013-10-25 2014-02-05 广东欧珀移动通信有限公司 照片处理方法和系统
CN103838484A (zh) * 2014-02-21 2014-06-04 联想(北京)有限公司 一种删除冗余图像的方法以及一种电子设备
CN104133917A (zh) * 2014-08-15 2014-11-05 百度在线网络技术（北京）有限公司 照片的分类存储方法及装置
CN104991959A (zh) * 2015-07-21 2015-10-21 北京京东尚科信息技术有限公司 一种基于内容检索相同或相似图像的方法与系统
CN105260385A (zh) * 2015-09-10 2016-01-20 上海斐讯数据通信技术有限公司 一种图片检索方法

Family Cites Families (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US9740963B2 (en) * 2014-08-05 2017-08-22 Sri International Multi-dimensional realization of visual content of an image collection

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20050105803A1 (en) * 2003-11-19 2005-05-19 Ray Lawrence A. Method for selecting an emphasis image from an image collection based upon content recognition
CN103558996A (zh) * 2013-10-25 2014-02-05 广东欧珀移动通信有限公司 照片处理方法和系统
CN103838484A (zh) * 2014-02-21 2014-06-04 联想(北京)有限公司 一种删除冗余图像的方法以及一种电子设备
CN104133917A (zh) * 2014-08-15 2014-11-05 百度在线网络技术（北京）有限公司 照片的分类存储方法及装置
CN104991959A (zh) * 2015-07-21 2015-10-21 北京京东尚科信息技术有限公司 一种基于内容检索相同或相似图像的方法与系统
CN105260385A (zh) * 2015-09-10 2016-01-20 上海斐讯数据通信技术有限公司 一种图片检索方法

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
See also references of EP3467677A4 *

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111274421A (zh) * 2020-01-15 2020-06-12 平安科技（深圳）有限公司 图片数据清洗方法、装置、计算机设备和存储介质
CN111274421B (zh) * 2020-01-15 2022-03-18 平安科技（深圳）有限公司 图片数据清洗方法、装置、计算机设备和存储介质

Also Published As

Publication number Publication date
EP3467677A1 (en) 2019-04-10
EP3467677A4 (en) 2019-12-11
CN107423297A (zh) 2017-12-01

Similar Documents

Publication Publication Date Title
WO2020224424A1 (zh) 图像处理方法、装置、计算机可读存储介质和计算机设备
CN103608826B (zh) 利用网络信息挖掘的视频内产品注释
WO2016054779A1 (en) Spatial pyramid pooling networks for image processing
CN107633023B (zh) 一种图像去重方法和装置
CN108875481B (zh) 用于行人检测的方法、装置、系统及存储介质
CN110765860A (zh) 摔倒判定方法、装置、计算机设备及存储介质
CN109871490B (zh) 媒体资源匹配方法、装置、存储介质和计算机设备
WO2017202086A1 (zh) 图片的筛选方法及装置
US20160125260A1 (en) Selecting features from image data
WO2021104125A1 (zh) 禽蛋异常的识别方法、装置及系统、存储介质、电子装置
CN111160288A (zh) 手势关键点检测方法、装置、计算机设备和存储介质
JP2006338313A (ja) 類似画像検索方法，類似画像検索システム，類似画像検索プログラム及び記録媒体
CN112434612A (zh) 吸烟检测方法、装置、电子设备及计算机可读存储介质
CN109815823B (zh) 数据处理方法及相关产品
CN109086690B (zh) 图像特征提取方法、目标识别方法及对应装置
CN103995864B (zh) 一种图像检索方法和装置
CN106096520A (zh) 一种信息推送方法及装置
CN111797971A (zh) 应用卷积神经网络进行数据处理的方法、装置和电子系统
CN108875505A (zh) 基于神经网络的行人再识别方法和装置
CN106611030B (zh) 基于视频的对象相似度比较方法和检索方法及其系统
Manh et al. Small object segmentation based on visual saliency in natural images
JP6132996B1 (ja) 画像処理装置，画像処理方法，画像処理プログラム
WO2024027347A9 (zh) 内容识别方法、装置、设备、存储介质及计算机程序产品
CN107688623A (zh) 一种实物检索方法、装置、设备及存储介质
US20220122341A1 (en) Target detection method and apparatus, electronic device, and computer storage medium

Legal Events

Date Code Title Description
NENP Non-entry into the national phase

Ref country code: DE

121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 17801940

Country of ref document: EP

Kind code of ref document: A1

ENP Entry into the national phase

Ref document number: 2017801940

Country of ref document: EP

Effective date: 20190102