WO2023071609A1 - 一种翻拍图像识别方法及其相关设备 - Google Patents

一种翻拍图像识别方法及其相关设备 Download PDF

Info

Publication number
WO2023071609A1
WO2023071609A1 PCT/CN2022/119635 CN2022119635W WO2023071609A1 WO 2023071609 A1 WO2023071609 A1 WO 2023071609A1 CN 2022119635 W CN2022119635 W CN 2022119635W WO 2023071609 A1 WO2023071609 A1 WO 2023071609A1
Authority
WO
WIPO (PCT)
Prior art keywords
image
feature
model
recognition
layer
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Ceased
Application number
PCT/CN2022/119635
Other languages
English (en)
French (fr)
Inventor
毛晓飞
黄灿
王长虎
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Beijing Youzhuju Network Technology Co Ltd
Original Assignee
Beijing Youzhuju Network Technology Co Ltd
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 Beijing Youzhuju Network Technology Co Ltd filed Critical Beijing Youzhuju Network Technology Co Ltd
Publication of WO2023071609A1 publication Critical patent/WO2023071609A1/zh
Anticipated expiration legal-status Critical
Ceased legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods

Definitions

  • the present application relates to the technical field of image processing, and in particular to a remake image recognition method and related equipment.
  • the application range of image uploading technology is becoming wider and wider.
  • the image uploading technology can be applied to some authentication scenarios (for example, real-name authentication, etc.), and can also be applied to some digital information extraction scenarios.
  • the image uploading technology it can not only perform uploading processing on conventional images, but also upload processing on reproduced images.
  • conventional image refers to the image data acquired by an image acquisition device (for example, a camera, a scanning device, etc.) for an image of a physical object.
  • “Copy image” refers to image data acquired by an image acquisition device for image data displayed on an image display device (for example, a display screen, etc.).
  • the present application provides a remake image recognition method and related equipment, which can accurately identify whether an image data is a remake image.
  • An embodiment of the present application provides a remake image recognition method, the method comprising:
  • the image to be identified is a duplicate image.
  • the visual feature includes a feature map to be used, and the feature map to be used includes image data of N channels; where N is a positive integer;
  • the determination process of the at least one feature to be used includes:
  • n is a positive integer, n ⁇ N;
  • n is a positive integer, n ⁇ N;
  • the at least one feature to be used is determined according to at least one extracted data segment of the N flattened data.
  • the method also includes:
  • the feature encoding processing is performed on each of the features to be used, and the result of feature encoding of each of the features to be used is obtained, including:
  • the determining the weighted weight value of each feature to be used includes:
  • Each of the features to be used is input into a pre-built weight determination model to obtain a weighted weight value of each of the features to be used output by the weight determination model.
  • the determining whether the image to be identified is a reproduced image according to the feature encoding result of the at least one feature to be used includes:
  • the recognition result of the image to be recognized it is determined whether the image to be recognized is a duplicate image.
  • performing recap image recognition processing on the encoded features of the image to be recognized to obtain a recognition result of the image to be recognized includes:
  • the construction process of the remake recognition model includes:
  • the constructing the remake recognition model by using the at least one image to be used and the actual label of the at least one image to be used includes:
  • the model to be trained is updated, and the step of inputting each of the images to be used into the model to be trained is continued until When the preset stop condition is reached, the remake recognition model is determined according to the model to be trained.
  • the model to be trained includes: a visual feature extraction layer, a feature splitting layer, a weight determination layer, a feature weighting layer, a feature encoding layer, a feature splicing layer, and a remake recognition layer;
  • the input data of the layer includes the output data of the feature splicing layer;
  • the input data of the feature splicing layer includes the output data of the feature encoding layer;
  • the input data of the feature encoding layer includes the output data of the feature weighting layer;
  • the input data of the feature weighting layer includes the output data of the weight determination layer and the output data of the feature split layer;
  • the input data of the weight determination layer includes the output data of the feature split layer;
  • the feature The input data of the split layer includes the output data of the visual feature extraction layer;
  • the determining the remake recognition model according to the model to be trained includes:
  • the remake recognition model is determined according to the remake recognition layer in the model to be trained.
  • the determining at least one image to be used according to the sample image includes:
  • At least one image to be used is determined.
  • the acquisition process of the image to be recognized includes:
  • the image captured by the device is acquired, the image captured by the device is adjusted according to a preset size to obtain the image to be recognized.
  • the embodiment of the present application also provides a duplicate image recognition device, including:
  • An extraction unit configured to perform visual feature extraction processing on the image to be identified after acquiring the image to be identified, to obtain visual features of the image to be identified;
  • a splitting unit configured to perform feature splitting processing on the visual features to obtain at least one feature to be used
  • An encoding unit configured to perform feature encoding processing on each of the features to be used, and obtain a feature encoding result of each of the features to be used;
  • An identifying unit configured to determine whether the image to be identified is a duplicate image according to a feature encoding result of the at least one feature to be used.
  • the embodiment of the present application also provides a device, the device includes a processor and a memory:
  • the memory is used to store computer programs
  • the processor is configured to execute any implementation manner of the remake image recognition method provided in the embodiments of the present application according to the computer program.
  • the embodiment of the present application also provides a computer-readable storage medium, the computer-readable storage medium is used to store a computer program, and the computer program is used to execute any implementation manner of the remake image recognition method provided in the embodiment of the present application .
  • the embodiment of the present application also provides a computer program product, which, when running on a terminal device, enables the terminal device to execute any implementation manner of the recapped image recognition method provided in the embodiment of the present application.
  • FIG. 1 is a flow chart of a remake image recognition method provided in an embodiment of the present application
  • FIG. 2 is a schematic structural diagram of a model to be trained provided in an embodiment of the present application
  • FIG. 3 is a schematic structural diagram of an apparatus for recognizing a duplicated image provided by an embodiment of the present application.
  • the inventor found that the moiré information carried by remake images is different from that carried by conventional images, so that there is a difference between the image information carried by the remake images and the image information carried by the conventional images, so Based on the image information carried by an image data, it can be identified whether the image data is a reproduced image.
  • the embodiment of the present application provides a remake image recognition method, the method includes: after the image to be recognized is acquired, visual feature extraction processing is performed on the image to be recognized , to obtain the visual features of the image to be recognized; then perform feature splitting on the visual features to obtain at least one feature to be used; then, perform feature encoding processing on each of the features to be used to obtain the features of each feature to be used Encoding result; finally, according to the feature encoding result of the at least one feature to be used, it is determined whether the image to be identified is a duplicate image.
  • the "at least one feature encoding result of the feature to be used” can accurately represent the image information carried by the image to be recognized
  • the "at least one feature coding result of the feature to be used” can accurately represent the image information carried by the image to be recognized.
  • Moiré information so that the "at least one feature encoding result of the feature to be used” can accurately indicate whether the image to be recognized is a remake image, so that the follow-up can be based on the "at least one feature encoding result of the feature to be used” , to identify whether an image data is a reproduced image.
  • the embodiment of the present application does not limit the subject of execution of the method for recaptured image recognition.
  • the image language recognition method provided in the embodiment of the present application can be applied to data processing devices such as terminal devices or servers.
  • the terminal device may be a smart phone, a computer, a personal digital assistant (Personal Digital Assistant, PDA), or a tablet computer.
  • the server can be an independent server, a cluster server or a cloud server.
  • FIG. 1 the figure is a flow chart of a remake image recognition method provided by an embodiment of the present application.
  • the remake image recognition method provided in the embodiment of this application includes S1-S4:
  • the "image to be recognized” refers to the image data that needs to be processed for image recognition processing; and the embodiment of the present application does not limit the acquisition process of the above "image to be recognized", for example, it can specifically be: after obtaining the image collected by the device , and directly determine the image collected by the device as the image to be recognized.
  • “Equipment-captured image” refers to image data collected by an image-capture device.
  • the embodiment of the present application also provides another possible implementation manner of acquiring the "image to be recognized", which may specifically include: after acquiring the image captured by the device, adjusting the image captured by the device according to a preset size, Get the image to be recognized.
  • the "preset size” may be preset; and this embodiment of the present application does not limit the "preset size", for example, it may specifically be 128 ⁇ 128.
  • Visual feature extraction processing is used to extract visual features from an image data; and the embodiment of the present application does not limit the implementation of the "visual feature extraction processing", for example, it can use any existing or future Visual feature extraction methods are implemented. As another example, it can be specifically implemented by means of a pre-built visual feature extraction model.
  • the above-mentioned "visual feature extraction model” is used to perform visual feature extraction processing on the input data of the visual feature extraction model; and the embodiment of the present application does not limit the “visual feature extraction model", for example, it can specifically be a convolutional neural network ( Convolutional Neural Networks, CNN) model.
  • CNN Convolutional Neural Networks
  • the embodiment of the present application does not limit the construction process of the "visual feature extraction model", for example, any existing or future model construction method can be used for implementation.
  • the model construction method shown in the third method embodiment below can also be used for implementation.
  • the "visual characteristics of the image to be recognized” is used to represent the image information carried by the image to be recognized; and the embodiment of the present application does not limit the acquisition process of the "visual characteristics of the image to be recognized", for example, it may specifically include:
  • the image is input into a pre-built visual feature extraction model, and the visual features of the image to be recognized outputted by the visual feature extraction model are obtained.
  • S2 Perform feature splitting on the visual features of the image to be recognized to obtain at least one feature to be used.
  • the “feature splitting process” is used to perform feature division processing for a visual feature (for example, a feature map); and the embodiment of the present application does not limit the implementation of the "feature splitting process", for example, the existing Or implement any method that can perform feature division processing on feature data in the future.
  • the embodiment of the present application also provides a possible implementation manner of "feature splitting processing".
  • feature splitting processing For ease of understanding, the following description will be made in conjunction with an example.
  • S2 may specifically include S21-S23:
  • S21 Perform flattening processing on the image data of the nth channel in the feature map to be used to obtain the nth flattened data.
  • n is a positive integer, n ⁇ N.
  • the "feature map to be used” is used to represent the image information carried by the image to be recognized; and the “feature map to be used” may include image data of N channels. Wherein, N is a positive integer.
  • this embodiment of the present application does not limit the "feature map to be used".
  • the “feature map to be used” may be 64 ⁇ 64 ⁇ 32 image data.
  • 32 refers to the number of channels of the “feature map to be used”
  • 64 ⁇ 64 refers to the data dimension of the image data of each channel in the "feature map to be used”.
  • the above-mentioned "flattening process” is used to expand and stitch an image data to obtain a data vector, so that the data vector is used to record the pixel information carried by the image data; and the embodiment of the present application does not limit the "flattening process".
  • treatment can be carried out by using any existing or future flattening treatment method.
  • nth flattened data refers to a data vector used to represent the image information carried by the image to be recognized; and this embodiment of the present application does not limit the “nth flattened data", for example, when the above "to be used When the image data of the nth channel in the feature map" is a 64 ⁇ 64-dimensional data matrix, the “nth flattened data” may be a 4096-dimensional data vector.
  • S22 Perform data extraction processing on the nth flattened data according to preset extraction parameters, to obtain at least one extracted data segment of the nth flattened data.
  • n is a positive integer, n ⁇ N.
  • the "preset extraction parameter” may be preset; and this embodiment of the present application does not limit the “preset extraction parameter”, for example, it may specifically be 1024 ⁇ 4.
  • “1024" refers to the data length of each extracted data segment; “4" refers to the number of extracted data segments.
  • data extraction processing is used to perform a data division operation on a data vector; and the embodiment of the present application does not limit the implementation of the "data extraction processing", for example, any existing or future emerging method that can A method of dividing a data vector into a plurality of extracted data segments (for example, a resize method) is implemented.
  • the above "at least one extracted data segment of the nth flattened data” is used to represent the image information carried by the nth flattened data; and this embodiment of the present application does not limit the "at least one extracted data segment of the nth flattened data” Segment", for example, when the nth flattened data is a 4096-dimensional data vector, and the preset extraction parameter is 1024 ⁇ 4, because the nth flattened data can be divided into four 1024-dimensional extracted data segments , so the "at least one extracted data segment of the nth flattened data” may include four 1024-dimensional extracted data segments.
  • S23 Determine at least one feature to be used according to at least one extracted data segment of the N pieces of flattened data.
  • the above-mentioned "at least one extracted data segment of the nth flattened data” includes four 1024-dimensional extracted data segments
  • the above-mentioned "at least one extracted data segment of the N flattened data” can Including 96 extracted data segments of 1024 dimensions, so S23 may specifically include: the first extracted data segment of 1024 dimensions, the second extracted data segment of 1024 dimensions, ..., and the 96th extracted data segment of 1024 dimensions All are determined as features to be used, so as to obtain 96 features to be used.
  • the visual features can be split into multiple feature data, so that the subsequent encoding process can be performed on the multiple feature data , to determine the encoding feature of the visual feature, which is conducive to improving the efficiency of subsequent encoding.
  • S3 Perform feature encoding processing on each feature to be used, and obtain a feature encoding result of each feature to be used.
  • the "feature encoding process” can be preset; and this embodiment of the present application does not limit the implementation of the "feature encoding process", for example, any existing or future encoding method can be used for implementation. As another example, it can be implemented with the help of a pre-built feature encoding model.
  • feature coding model is used to encode the input data of the feature coding model; and the embodiment of the present application does not limit the model structure of the "feature coding model", for example, it may include 5 coding networks. It should be noted that the embodiment of the present application does not limit the implementation of the "encoding network", for example, any encoding method (for example, the Encoder module in the transformer model, or the conformer structure) can be used for implementation.
  • the embodiment of the present application does not limit the construction method of the "feature encoding model", for example, any existing or future model construction method can be used for implementation.
  • the model construction method shown in the third method embodiment below can also be used for implementation.
  • the feature encoding result of the kth feature to be used is used to represent the image information (for example, moiré information) carried by the kth feature to be used; and the embodiment of the present application does not limit the feature of the "kth feature to be used
  • the process of determining the "coding result" may specifically be: input the kth feature to be used into a pre-built feature coding model, and obtain the feature coding result of the kth feature to be used output by the feature coding model.
  • k is a positive integer
  • k ⁇ K K is a positive integer
  • K represents the number of features to be used (for example, 96).
  • the embodiment of the present application does not limit the implementation of S4.
  • it may specifically be: input the feature encoding result of at least one feature to be used into a pre-trained remake image detection model, and obtain the detection result output by the remake image detection model, and use The detection result is used to indicate whether the image to be identified is a duplicate image.
  • remake image detection model is used to perform remake image detection and recognition processing for the input data of the remake image detection model; and the embodiment of the present application does not limit the "remake image detection model", for example, it may be an existing or future Any machine learning model that emerges. It should be noted that the embodiment of the present application does not limit the training process of the "remake image detection model", for example, any existing or future model training method can be used for implementation.
  • S41 Concatenate the feature coding results of at least one feature to be used to obtain the coding feature of the image to be recognized.
  • the "coding feature of the image to be recognized” is used to represent the image information carried by the image to be recognized (especially, moiré information, etc.).
  • the embodiment of the present application does not limit the implementation of the above "splicing", and any existing or future splicing method can be used for implementation.
  • S42 Perform image recognition processing on recapped images based on the encoded features of the image to be recognized to obtain a recognition result of the image to be recognized.
  • the "remake image recognition processing” is used to identify whether the coding feature of an image data belongs to the characteristics of a remake image; and the embodiment of the present application does not limit the “remake image recognition processing", for example, it can use Identify the model for implementation.
  • remake recognition model is used to perform remake image recognition processing for the input data of the remake recognition model; and the embodiment of the present application does not limit the model structure of the "remake recognition model", for example, it may include a linear processing layer and a fully connected layer, and the input data of the fully connected layer includes the output data of the linear processing layer.
  • the embodiment of the present application does not limit the implementation of the above-mentioned "linear processing layer", for example, it can be implemented by using any existing or future linear processing method (for example, the linear module in the transformer model).
  • the embodiment of the present application does not limit the implementation of the above-mentioned "fully connected layer”, for example, it can be implemented by using any existing or future fully connected processing method (for example, the softmax module in the transformer model).
  • the embodiment of the present application does not limit the construction method of the "remake recognition model", for example, any existing or future model construction method can be used for implementation.
  • the model construction method shown in the third method embodiment below can also be used for implementation.
  • the above-mentioned "recognition result of the image to be recognized” is used to indicate the possibility that the image to be recognized belongs to a duplicate image; and the embodiment of the present application does not limit the expression of the "recognition result of the image to be recognized", for example, it may include the first probability and second probability.
  • the "first probability” is used to indicate the possibility that the image to be recognized is a remake; the “second probability” is used to indicate the possibility that the image to be recognized is not a remake.
  • the embodiment of the present application does not limit the implementation of S42.
  • it may specifically include: inputting the coding features of the image to be recognized into a pre-built remake recognition model, and obtaining the recognition result of the image to be recognized output by the remake recognition model.
  • the "preset remake condition” may be preset, for example, the “preset remake condition” may specifically be: the first probability is higher than the first threshold, and/or the second probability is lower than the second threshold.
  • the image to be recognized is firstly subjected to visual feature extraction processing to obtain the image to be recognized Visual features; then perform feature splitting processing on the visual features to obtain at least one feature to be used; then, perform feature encoding processing on each of the features to be used to obtain the feature encoding results of each feature to be used; finally, according to the A feature encoding result of at least one feature to be used is used to determine whether the image to be identified is a duplicate image.
  • the "at least one feature encoding result of the feature to be used” can accurately represent the image information carried by the image to be recognized
  • the "at least one feature coding result of the feature to be used” can accurately represent the image information carried by the image to be recognized.
  • Moiré information so that the "at least one feature encoding result of the feature to be used” can accurately indicate whether the image to be recognized is a remake image, so that the follow-up can be based on the "at least one feature encoding result of the feature to be used” , to identify whether an image data is a reproduced image.
  • the embodiment of the present application also provides another possible implementation of the recaptured image recognition method.
  • the recaptured image recognition method includes: Can also include S5-S7:
  • the weighted weight value of the kth feature to be used is used to represent the importance of the image information represented by the kth feature to be used; and the embodiment of the present application does not limit the "weighted weight value of the kth feature to be used
  • the way of obtaining ", for example, can be preset.
  • k is a positive integer
  • k ⁇ K K is a positive integer
  • K represents the number of features to be used (for example, 96).
  • the embodiment of the present application also provides another possible implementation of S5, which may specifically include: inputting each feature to be used into a pre-built weight determination model to obtain the weight determination model The weighted weight value of each feature output by the model to be used.
  • weight determination model is used to measure the importance of information for the input data of the weight determination model; and the embodiment of the present application does not limit the “weight determination model", for example, it can be any machine learning model (such as , 1 ⁇ 1 convolutional network) is enough.
  • the embodiment of the present application does not limit the construction process of the above "weight determination model", for example, it can be implemented by using any existing or future model construction method.
  • the model construction method shown in the third method embodiment below can also be used for implementation.
  • S6 may specifically include: multiplying the k-th feature to be used by the weighted weight value of the k-th feature to be used to obtain the k-th weighted feature, so that the k-th weighted feature can be more accurately Indicates the image information carried by the k-th feature to be used and the degree of influence it has on subsequent image reproduction recognition processing.
  • S7 Perform feature encoding processing on each weighted feature respectively to obtain a feature encoding result of each feature to be used.
  • At least one feature to be used of the image to be identified can be referred to and its weighted weight value, and at least one feature to be used of the image to be identified can be encoded.
  • the embodiment of the present application also provides a model construction method, which may specifically include step 11-step 14:
  • Step 11 Obtain a sample image and the actual label of the sample image.
  • sample image refers to the image data that needs to be used in the model building process; and the embodiment of the present application does not limit the number of “sample images”.
  • actual label of the sample image is used to indicate whether the sample image is actually a retake image; and the embodiment of the present application does not limit the acquisition process of the "actual label of the sample image", for example, it can be implemented by manual labeling.
  • this embodiment of the present application does not limit the expression manner of the above-mentioned "actual label of the sample image", for example, it may include a third probability and a fourth probability.
  • the "third probability” is used to indicate the possibility that the sample image actually belongs to a remake image
  • the "fourth probability” is used to indicate the possibility that the sample image does not actually belong to a remake image.
  • Step 12 Determine at least one image to be used according to the sample image.
  • image to be used is used to represent part of image information or all image information in the sample image.
  • this embodiment of the present application does not limit the above "at least one image to be used", for example, it may include one first image sample, H second image samples, and G third image samples.
  • H is a positive integer
  • first image sample is used to record all image information in the sample image; and this embodiment of the present application does not limit the determination process of the "first image sample", for example, the sample image can be directly determined as the first image sample.
  • the above-mentioned “second image sample” is used to record the image information of the sample image under the first cropping ratio (for example, 70%); and the embodiment of the present application does not limit the determination process of the above-mentioned "H second image samples", for example,
  • the sample image may be randomly cropped h times according to the first cropping ratio to obtain the h second image sample, so that the image information carried by the h second image sample is different from the image information carried by the sample image The same ratio between achieve the "first crop ratio" mentioned above.
  • the above-mentioned "third image sample” is used to record the image information in the sample image under the second cropping ratio (for example, 40%); and the embodiment of the present application does not limit the determination process of the above-mentioned "G third image samples", for example,
  • the sample image may be randomly cropped for the gth time according to the second cropping ratio to obtain the gth third image sample, so that the image information carried by the gth third image sample is different from the image information carried by the sample image The same ratio between achieve the "second crop ratio" mentioned above.
  • step 12 does not limit the implementation of step 12, for example, it may specifically include step 121-step 122:
  • Step 121 Perform cropping processing on the sample image according to preset rules to obtain at least one cropped image.
  • the preset rule may be set in advance, for example, it may specifically include: performing H cropping processing on the sample image according to the first cropping ratio, and performing G cropping processing on the sample image according to the second cropping ratio.
  • the above "cropped image” is used to represent part of the image information in the sample image.
  • Step 122 Determine at least one image to be used according to the sample image and at least one cropped image.
  • step 122 does not limit the implementation manner of step 122.
  • it may specifically be: directly determining the sample image and each cropped image as images to be used.
  • step 122 may specifically be: firstly perform data filling processing on each cropped image according to the image size of the sample image to obtain each filled image, so that the image size of each filled image is consistent with the image size of the sample image; The sample image and each filling image are determined as images to be used.
  • the image information of different regions in the sample image can be referred to to determine different images to be used, so that the different images to be used can represent the images carried by different regions in the sample image.
  • Image information so that each model constructed based on these images to be used has a better image region feature extraction performance, which is conducive to improving the feature extraction effect, thereby helping to improve the recognition effect of the reproduced image.
  • Step 13 Determine the actual label of the sample image as the actual label of each image to be used.
  • the actual label of each image to be used can be determined according to the actual label of the sample image, so that the actual label of each image to be used is consistent with the actual label of the sample image. unanimous.
  • Step 14 Using at least one image to be used and the actual label of the at least one image to be used, construct at least one of a visual feature extraction model, a feature encoding model, a weight determination model, and a remake recognition model.
  • step 14 may specifically include step 141-step 144:
  • Step 141 Input each image to be used into the model to be trained, and obtain the recognition result of each image to be used output by the model to be trained.
  • model to be trained is used to perform recapture image recognition processing on the input data of the model to be trained; and this embodiment of the present application does not limit the "model to be trained", for example, it may be a machine learning model.
  • the embodiment of the present application does not limit the model structure of the above-mentioned "model to be trained".
  • the input data of the remake recognition layer includes the output data of the feature splicing layer;
  • the input data of the feature splicing layer includes the output data of the feature encoding layer;
  • the input data of the feature encoding layer includes the output data of the feature weighting layer;
  • the input data of the feature weighting layer It includes the output data of the weight determination layer and the output data of the feature split layer;
  • the input data of the weight determination layer includes the output data of the feature split layer;
  • the input data of the feature split layer includes the output data of the visual feature extraction layer.
  • the above-mentioned "visual feature extraction layer” is used to perform visual feature extraction processing for an image data; and the embodiment of the present application does not limit the network structure of the "visual feature extraction layer", for example, it can use the above “visual feature extraction model” The model structure is implemented.
  • feature splitting layer is used to perform feature splitting processing on the input data of the feature splitting layer; and the embodiment of the present application does not limit the working principle of the "feature splitting layer", for example, it can adopt the above S21 - The feature splitting process shown in S23 is implemented.
  • weight determination layer is used to measure the importance of information for the input data of the weight determination layer; and the embodiment of the present application does not limit the network structure of the "weight determination layer", for example, it can adopt the above "weight determination layer” Model” model structure for implementation.
  • feature weighting layer is used to perform weighting processing on the input data of the weight determination layer; and the embodiment of the present application does not limit the working principle of the "feature weighting layer", for example, it can adopt the weighting processing shown in S6 above process is implemented.
  • feature encoding layer is used to encode the input data of the feature encoding layer; and the embodiment of the present application does not limit the network structure of the "feature encoding layer", for example, it can adopt the above “feature encoding model”
  • the model structure is implemented.
  • feature splicing layer is used to splice the input data of the feature splicing layer; and the embodiment of the present application does not limit the working principle of the "feature splicing layer", and any existing or future splicing can be used method is implemented.
  • remake recognition layer is used to perform remake data recognition processing for the input data of the remake recognition layer; and the embodiment of the present application does not limit the network structure of the "remake recognition layer", for example, it can use the above "remake recognition model " model structure for implementation.
  • the aforementioned "recognition result of the image to be used" is used to indicate the possibility that the image to be used is predicted to belong to a retake image.
  • Step 142 Determine whether the preset stop condition is met, if yes, execute step 144; if not, execute step 143.
  • the "preset stop condition" can be preset; and the embodiment of the present application does not limit the “preset stop condition”, for example, it can specifically be that the model loss value of the model to be trained is lower than the preset loss value threshold; It may be that the rate of change of the model loss value of the model to be trained is lower than a preset rate of change threshold; it may also be that the number of updates of the model to be trained reaches a preset threshold.
  • model loss value of the model to be trained is used to represent the recap image recognition performance of the model to be trained; and the embodiment of the present application does not limit the determination process of the "model loss value of the model to be trained", existing or Any model loss value calculation method that will appear in the future will be implemented.
  • Step 143 Update the model to be trained according to the recognition result of the at least one image to be used and the actual label of the at least one image to be used, and return to step 141.
  • the remake image recognition performance of the model to be trained is still relatively poor, so it can be based on the recognition result of at least one image to be used and the The difference between the actual labels of at least one image to be used, update the model to be trained, so that the updated model to be trained has better remake image recognition performance, and continue to execute steps 141 and 2 based on the updated model to be trained Its subsequent steps are to realize the next round of training process for the model to be trained.
  • Step 144 According to the model to be trained, determine at least one of a visual feature extraction model, a feature encoding model, a weight determination model, and a remake recognition model.
  • the visual feature extraction model can be determined directly based on the model to be trained , feature encoding model, weight determination model, and at least one of the remake recognition model; and the determination process may specifically be: according to the visual feature extraction layer in the model to be trained, determine the visual feature extraction model (for example, directly the to-be-trained The visual feature extraction layer in the model is determined as the visual feature extraction model); according to the feature coding layer in the model to be trained, the feature coding model is determined (for example, the feature coding layer in the model to be trained is directly determined as the feature coding model); According to the weight determination layer in the model to be trained, determine the weight determination model (for example, directly determine the weight determination layer in the model to be trained as the weight determination model); according to the remake recognition layer in the model to be trained, determine the remake recognition model ( For example, the remake recognition
  • the embodiment of the present application also provides a duplicated image recognition device, which will be explained and described below with reference to the accompanying drawings.
  • this figure is a schematic structural diagram of an apparatus for recognizing a duplicated image provided by an embodiment of the present application.
  • the reproduced image recognition device 300 provided in the embodiment of the present application includes:
  • the extraction unit 301 is configured to, after the image to be recognized is acquired, perform visual feature extraction processing on the image to be recognized to obtain visual features of the image to be recognized;
  • An encoding unit 303 configured to perform feature encoding processing on each of the features to be used, and obtain a feature encoding result of each of the features to be used;
  • the identifying unit 304 is configured to determine whether the image to be identified is a retake image according to a feature encoding result of the at least one feature to be used.
  • the visual feature includes a feature map to be used, and the feature map to be used includes image data of N channels; where N is a positive integer;
  • the splitting unit 302 is specifically configured to: flatten the image data of the nth channel in the feature map to be used to obtain the nth flattened data; wherein, n is a positive integer, n ⁇ N; Performing data extraction processing on the nth flattened data according to preset extraction parameters to obtain at least one extracted data segment of the nth flattened data; wherein, n is a positive integer, n ⁇ N; according to the Nth flattened data At least one extracted data segment of flat data, and the at least one feature to be used is determined.
  • the reproduced image recognition device 300 further includes:
  • a determining unit configured to determine the weighted weight value of each of the features to be used
  • a weighting unit configured to multiply each of the features to be used by the weighted weight value of each of the features to be used to obtain each weighted feature
  • the encoding unit 303 is specifically configured to: perform feature encoding processing on each of the weighted features to obtain a feature encoding result of each of the features to be used.
  • the determining unit is specifically configured to: input each of the features to be used into a pre-built weight determination model, and obtain the weighted weight of each of the features to be used output by the weight determination model value.
  • the identification unit 304 includes:
  • a splicing subunit configured to splice the feature coding results of the at least one feature to be used to obtain the coding features of the image to be recognized;
  • a processing subunit configured to perform recapture image recognition processing on the coding features of the image to be recognized to obtain a recognition result of the image to be recognized;
  • the determination subunit is configured to determine whether the image to be identified is a duplicate image according to the identification result of the image to be identified.
  • the processing subunit is specifically configured to: input the coding features of the image to be recognized into a pre-built remake recognition model, and obtain the output of the image to be recognized by the remake recognition model. recognition result.
  • the reproduced image recognition device 300 further includes:
  • a construction unit configured to acquire a sample image and an actual label of the sample image; wherein, the actual label of the sample image is used to indicate whether the sample image is actually a remake image; according to the sample image, determine at least one to-be-used image; determining the actual label of the sample image as the actual label of each image to be used; using the at least one image to be used and the actual label of the at least one image to be used to construct the remake recognition model.
  • the construction unit includes:
  • An identification subunit configured to input each of the images to be used into the model to be trained, and obtain a recognition result of each of the images to be used output by the model to be trained;
  • the update subunit is used to update the model to be trained according to the recognition result of the at least one image to be used and the actual label of the at least one image to be used, and return to the recognition subunit to continue performing the step of converting each of the images to be used. Describe the steps to use the image input model to be trained;
  • a subunit is constructed for determining the remake recognition model according to the model to be trained when a preset stop condition is reached.
  • the model to be trained includes: a visual feature extraction layer, a feature splitting layer, a weight determination layer, a feature weighting layer, a feature encoding layer, a feature splicing layer, and a remake recognition layer;
  • the input data of the layer includes the output data of the feature splicing layer;
  • the input data of the feature splicing layer includes the output data of the feature encoding layer;
  • the input data of the feature encoding layer includes the output data of the feature weighting layer;
  • the input data of the feature weighting layer includes the output data of the weight determination layer and the output data of the feature split layer;
  • the input data of the weight determination layer includes the output data of the feature split layer;
  • the feature The input data of the split layer includes the output data of the visual feature extraction layer;
  • the construction subunit is specifically configured to: determine the remake recognition model according to the remake recognition layer in the model to be trained.
  • the splitting unit 302 is specifically configured to: perform cropping processing on the sample image according to preset rules to obtain at least one cropped image; according to the sample image and the at least one Crop the image and determine at least one image to be used.
  • the reproduced image recognition device 300 further includes:
  • the adjusting unit is configured to, after acquiring the image captured by the device, adjust the image captured by the device according to a preset size to obtain the image to be recognized.
  • the remake image recognition device 300 Based on the relevant content of the remake image recognition device 300, it can be known that for the remake image recognition device 300, after the image to be recognized is acquired, the image to be recognized is first subjected to visual feature extraction processing to obtain the visual features of the image to be recognized; Then, perform feature splitting processing on the visual feature to obtain at least one feature to be used; then, perform feature encoding processing on each feature to be used respectively to obtain the feature encoding result of each feature to be used; finally, according to the at least one feature to be used Using the feature encoding result of the feature, it is determined whether the image to be identified is a duplicate image.
  • the "at least one feature encoding result of the feature to be used” can accurately represent the image information carried by the image to be recognized
  • the "at least one feature coding result of the feature to be used” can accurately represent the image information carried by the image to be recognized Moiré information, so that the "at least one feature encoding result of the feature to be used” can accurately indicate whether the image to be recognized is a remake image, so that the follow-up can be based on the "at least one feature encoding result of the feature to be used” , to identify whether an image data is a reproduced image.
  • the embodiment of the present application also provides a device, the device includes a processor and a memory:
  • the memory is used to store computer programs
  • the processor is configured to execute any implementation manner of the remake image recognition method provided in the embodiments of the present application according to the computer program.
  • the embodiment of the present application also provides a computer-readable storage medium, the computer-readable storage medium is used to store a computer program, and the computer program is used to execute any of the remake image recognition methods provided in the embodiment of the present application.
  • a computer-readable storage medium is used to store a computer program
  • the computer program is used to execute any of the remake image recognition methods provided in the embodiment of the present application.
  • an embodiment of the present application also provides a computer program product, which, when run on a terminal device, enables the terminal device to execute any implementation manner of the duplicated image recognition method provided in the embodiment of the present application.
  • At least one (item) means one or more, and “multiple” means two or more.
  • “And/or” is used to describe the association relationship of associated objects, indicating that there can be three types of relationships, for example, “A and/or B” can mean: only A exists, only B exists, and A and B exist at the same time , where A and B can be singular or plural.
  • the character “/” generally indicates that the contextual objects are an “or” relationship.
  • At least one of the following” or similar expressions refer to any combination of these items, including any combination of single or plural items.
  • At least one item (piece) of a, b or c can mean: a, b, c, "a and b", “a and c", “b and c", or "a and b and c ", where a, b, c can be single or multiple.

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • Evolutionary Computation (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Artificial Intelligence (AREA)
  • General Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • General Health & Medical Sciences (AREA)
  • Software Systems (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • Biophysics (AREA)
  • Biomedical Technology (AREA)
  • Mathematical Physics (AREA)
  • Computational Linguistics (AREA)
  • Health & Medical Sciences (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Evolutionary Biology (AREA)
  • Image Analysis (AREA)

Abstract

本申请公开了一种翻拍图像识别方法及其相关设备，该方法包括：在获取到待识别图像之后，先对该待识别图像进行视觉特征提取处理，得到该待识别图像的视觉特征；再对该视觉特征进行特征拆分处理，得到至少一个待使用特征；然后，分别对各个该待使用特征进行特征编码处理，得到各个该待使用特征的特征编码结果；最后，根据该至少一个待使用特征的特征编码结果，确定该待识别图像是否为翻拍图像。其中，因上述"至少一个待使用特征的特征编码结果"能够准确地表示出待识别图像携带的摩尔纹信息，使得其能够准确地表示出该待识别图像是否为翻拍图像，如此使得后续能够基于该"至少一个待使用特征的特征编码结果"，识别出一个图像数据是否为翻拍图像。

Description

一种翻拍图像识别方法及其相关设备
本公开要求于2021年10月29日提交中国专利局、申请号为202111275642.9、发明名称为“一种翻拍图像识别方法及其相关设备”的中国专利申请的优先权，其全部内容通过引用结合在本公开中。
技术领域
本申请涉及图像处理技术领域，尤其涉及一种翻拍图像识别方法及其相关设备。
背景技术
随着信息化技术的普及，图像上传技术的应用范围越来越广。例如，图像上传技术可以应用于一些认证场景(例如，实名认证等)，也可以应用于一些数字信息提取场景。
另外，对于图像上传技术来说，其不仅可以针对常规图像进行上传处理，也可以针对翻拍图像进行上传处理。其中，“常规图像”是指由图像采集设备(例如，摄像头、扫描设备等)针对实体对象进行图像采集得到的图像数据。“翻拍图像”是指由图像采集设备针对图像显示设备(例如，显示屏等)上显示的图像数据进行图像采集得到的图像数据。
然而，对于一些应用场景来说，常规图像通常是一种合法图像数据，但是翻拍图像通常是一种非法图像数据，故为了提高用户信息安全，需要针对翻拍图像进行识别并告警，如此使得如何识别翻拍图像成为一个亟待解决的技术问题。
发明内容
为了解决上述技术问题，本申请提供了一种翻拍图像识别方法及其相关设备，能够准确地识别出一个图像数据是否为翻拍图像。
为了实现上述目的，本申请实施例提供的技术方案如下：
本申请实施例提供一种翻拍图像识别方法，所述方法包括：
在获取到待识别图像之后，对所述待识别图像进行视觉特征提取处理，得到所述待识别图像的视觉特征；
对所述视觉特征进行特征拆分处理，得到至少一个待使用特征；
分别对各个所述待使用特征进行特征编码处理，得到各个所述待使用特征的特征编码结果；
根据所述至少一个待使用特征的特征编码结果，确定所述待识别图像是否为翻拍图像。
在一种可能的实施方式中，所述视觉特征包括待使用特征图，且所述待使用特征图包括N个通道的图像数据；其中，N为正整数；
所述至少一个待使用特征的确定过程，包括：
对所述待使用特征图中第n个通道的图像数据进行压平处理，得到第n个压平数据；其中，n为正整数，n≤N；
将所述第n个压平数据按照预设提取参数进行数据提取处理，得到所述第n个压平数据的至少一个提取数据段；其中，n为正整数，n≤N；
根据N个压平数据的至少一个提取数据段，确定所述至少一个待使用特征。
在一种可能的实施方式中，所述方法还包括：
确定各个所述待使用特征的加权权重值；
将各个所述待使用特征分别与各个所述待使用特征的加权权重值进行相乘，得到各个加权特征；
所述分别对各个所述待使用特征进行特征编码处理，得到各个所述待使用特征的特征编码结果，包括：
分别对各个所述加权特征进行特征编码处理，得到各个所述待使用特征的特征编码结果。
在一种可能的实施方式中，所述确定各个所述待使用特征的加权权重值，包括：
将各个所述待使用特征输入预先构建的权重确定模型，得到所述权重确定模型输出的各个所述待使用特征的加权权重值。
在一种可能的实施方式中，所述根据所述至少一个待使用特征的特征编码结果，确定所述待识别图像是否为翻拍图像，包括：
将所述至少一个待使用特征的特征编码结果进行拼接，得到所述待识别图像的编码特征；
对所述待识别图像的编码特征进行翻拍图像识别处理，得到所述待识别图像的识别结果；
根据所述待识别图像的识别结果，确定所述待识别图像是否为翻拍图像。
在一种可能的实施方式中，所述对所述待识别图像的编码特征进行翻拍图像识别处理，得到所述待识别图像的识别结果，包括：
将所述待识别图像的编码特征输入预先构建的翻拍识别模型，得到所述翻拍识别模型输出的所述待识别图像的识别结果。
在一种可能的实施方式中，所述翻拍识别模型的构建过程，包括：
获取样本图像和所述样本图像的实际标签；其中，所述样本图像的实际标签用于表示所述样本图像实际是否为翻拍图像；
根据所述样本图像，确定至少一个待使用图像；
将所述样本图像的实际标签确定为各个待使用图像的实际标签；
利用所述至少一个待使用图像和所述至少一个待使用图像的实际标签，构建所述翻拍识别模型。
在一种可能的实施方式中，所述利用所述至少一个待使用图像和所述至少一个待使用图像的实际标签，构建所述翻拍识别模型，包括：
将各个所述待使用图像输入待训练模型，得到所述待训练模型输出的各个所述待使用图像的识别结果；
根据所述至少一个待使用图像的识别结果和所述至少一个待使用图像的实际标签，更新所述待训练模型，并继续执行所述将各个所述待使用图像输入待训练模型的步骤，直至在达到预设停止条件时，根据所述待训练模型，确定所述翻拍识别模型。
在一种可能的实施方式中，所述待训练模型包括：视觉特征提取层、特征拆分层、权重确定层、特征加权层、特征编码层、特征拼接层和翻拍识别层；所述翻拍识别层的输入数据包括所述特征拼接层的输出数据；所述特征拼接层的输入数据包括所述特征编码层的输出数据；所述特征编码层的输入数据包括所述特征加权层的输出数据；所述特征加权层的输入数据包括所述权重确定层的输出数据和所述特征拆分层的输出数据；所述权重确定层的输入数据包括所述特征拆分层的输出数据；所述特征拆分层的输入数据包括所述视觉特征提取层的输出数据；
所述根据所述待训练模型，确定所述翻拍识别模型，包括：
根据所述待训练模型中翻拍识别层，确定所述翻拍识别模型。
在一种可能的实施方式中，所述根据所述样本图像，确定至少一个待使用图像，包括：
按照预设规则，对所述样本图像进行裁剪处理，得到至少一个裁剪图像；
根据所述样本图像和所述至少一个裁剪图像，确定至少一个待使用图像。
在一种可能的实施方式中，所述待识别图像的获取过程，包括：
在获取到设备采集图像之后，对所述设备采集图像按照预设尺寸进行调整处理，得到所述待识别图像。
本申请实施例还提供了一种翻拍图像识别装置，包括：
提取单元，用于在获取到待识别图像之后，对所述待识别图像进行视觉特征提取处理，得到所述待识别图像的视觉特征；
拆分单元，用于对所述视觉特征进行特征拆分处理，得到至少一个待使用特征；
编码单元，用于分别对各个所述待使用特征进行特征编码处理，得到各个所述待使用特征的特征编码结果；
识别单元，用于根据所述至少一个待使用特征的特征编码结果，确定所述待识别图像是否为翻拍图像。
本申请实施例还提供了一种设备，所述设备包括处理器以及存储器：
所述存储器用于存储计算机程序；
所述处理器用于根据所述计算机程序执行本申请实施例提供的翻拍图像识别方法的任一实施方式。
本申请实施例还提供了一种计算机可读存储介质，所述计算机可读存储介质用于存储计算机程序，所述计算机程序用于执行本申请实施例提供的翻拍图像识别方法的任一实施方式。
本申请实施例还提供了一种计算机程序产品，所述计算机程序产品在终端设备上运行时，使得所述终端设备执行本申请实施例提供的翻拍图像识别方法的任一实施方式。
附图说明
为了更清楚地说明本申请实施例或现有技术中的技术方案，下面将对实施例或现有技术描述中所需要使用的附图作简单地介绍，显而易见地，下面描述中的附图仅仅是本申请中记载的一些实施例，对于本领域普通技术人员来讲，在不付出创造性劳动的前提下，还 可以根据这些附图获得其它的附图。
图1为本申请实施例提供的一种翻拍图像识别方法的流程图；
图2为本申请实施例提供的一种待训练模型的结构示意图；
图3为本申请实施例提供的一种翻拍图像识别装置的结构示意图。
具体实施方式
发明人在针对翻拍图像的研究中发现，翻拍图像携带的摩尔纹信息不同于常规图像携带的摩尔纹信息，使得该翻拍图像携带的图像信息与该常规图像携带的图像信息之间存在区别，故可以基于一个图像数据携带的图像信息，识别该图像数据是否为翻拍图像。
基于上述发现可知，为了解决背景技术部分的技术问题，本申请实施例提供了一种翻拍图像识别方法，该方法包括：在获取到待识别图像之后，先对该待识别图像进行视觉特征提取处理，得到该待识别图像的视觉特征；再对该视觉特征进行特征拆分处理，得到至少一个待使用特征；然后，分别对各个该待使用特征进行特征编码处理，得到各个该待使用特征的特征编码结果；最后，根据该至少一个待使用特征的特征编码结果，确定该待识别图像是否为翻拍图像。
可见，因上述“至少一个待使用特征的特征编码结果”能够准确地表示出待识别图像携带的图像信息，使得该“至少一个待使用特征的特征编码结果”能够准确地表示出待识别图像携带的摩尔纹信息，从而使得该“至少一个待使用特征的特征编码结果”能够准确地表示出该待识别图像是否为翻拍图像，如此使得后续能够基于该“至少一个待使用特征的特征编码结果”，识别出一个图像数据是否为翻拍图像。
另外，本申请实施例不限定翻拍图像识别方法的执行主体，例如，本申请实施例提供的图像语种识别方法可以应用于终端设备或服务器等数据处理设备。其中，终端设备可以为智能手机、计算机、个人数字助理(Personal Digital Assitant，PDA)或平板电脑等。服务器可以为独立服务器、集群服务器或云服务器。
为了使本技术领域的人员更好地理解本申请方案，下面将结合本申请实施例中的附图，对本申请实施例中的技术方案进行清楚、完整地描述，显然，所描述的实施例仅是本申请一部分实施例，而不是全部的实施例。基于本申请中的实施例，本领域普通技术人员在没有做出创造性劳动前提下所获得的所有其他实施例，都属于本申请保护的范围。
方法实施例一
参见图1，该图为本申请实施例提供的一种翻拍图像识别方法的流程图。
本申请实施例提供的翻拍图像识别方法，包括S1-S4：
S1：在获取到待识别图像之后，对该待识别图像进行视觉特征提取处理，得到该待识别图像的视觉特征。
其中，“待识别图像”是指需要进行翻拍图像识别处理的图像数据；而且本申请实施例不限定上述“待识别图像”的获取过程，例如，其具体可以为：在获取到设备采集图像之后，直接将该设备采集图像确定为待识别图像。“设备采集图像”是指由图像采集设备采集到的图像数据。
另外，本申请实施例还提供了获取“待识别图像”的另一种可能的实施方式，其具体可以包括：在获取到设备采集图像之后，对该设备采集图像按照预设尺寸进行调整处理，得到待识别图像。其中，“预设尺寸”可以预先设定；而且本申请实施例不限定该“预设尺寸”，例如，其具体可以为128×128。
需要说明的是，本申请实施例不限定上述“调整处理”的实施方式，例如，可以采用现有的或者未来出现的任一种能够针对一个图像数据进行尺寸调整的方法(例如，尺寸调整(resize)方法)进行实施。
“视觉特征提取处理”用于从一个图像数据中提取视觉特征；而且本申请实施例不限定该“视觉特征提取处理”的实施方式，例如，其可以采用现有的或者未来出现的任一种视觉特征提取方法进行实施。又如，其具体可以借助预先构建的视觉特征提取模型进行实施。
上述“视觉特征提取模型”用于针对该视觉特征提取模型的输入数据进行视觉特征提取处理；而且本申请实施例不限定该“视觉特征提取模型”，例如，其具体可以为卷积神经网络(Convolutional Neural Networks,CNN)模型。
另外，本申请实施例不限定“视觉特征提取模型”的构建过程，例如，可以采用现有的或者未来出现的任一种模型构建方法进行实施。又如，还可以采用下文 方法实施例三所示的模型构建方法进行实施。
“待识别图像的视觉特征”用于表示该待识别图像携带的图像信息；而且本申请实施例不限定该“待识别图像的视觉特征”的获取过程，例如，其具体可以包括：将待识别图像输入预先构建的视觉特征提取模型，得到该视觉特征提取模型输出的该待识别图像的视觉特征。
S2：对待识别图像的视觉特征进行特征拆分处理，得到至少一个待使用特征。
其中，“特征拆分处理”用于针对一个视觉特征(例如，特征图)进行特征划分处理；而且本申请实施例不限定该“特征拆分处理”的实施方式，例如，可以采用现有的或者未来出现的任一种能够针对特征数据进行特征划分处理的方法进行实施。
另外，本申请实施例还提供了“特征拆分处理”的一种可能的实施方式，为了便于理解，下面结合示例进行说明。
作为示例，当上述“待识别图像的视觉特征”包括待使用特征图，且该待使用特征图包括N个通道的图像数据时，则S2具体可以包括S21-S23：
S21：对待使用特征图中第n个通道的图像数据进行压平处理，得到第n个压平数据。其中，n为正整数，n≤N。
其中，“待使用特征图”用于表示待识别图像携带的图像信息；而且该“待使用特征图”可以包括N个通道的图像数据。其中，N为正整数。
另外，本申请实施例不限定该“待使用特征图”，例如，当上述N为32时，该“待使用特征图”可以是64×64×32的图像数据。其中，“32”是指“待使用特征图”的通道个数；“64×64”是指“待使用特征图”中各个通道的图像数据所具有的数据维度。
上述“压平处理”用于将一个图像数据进行展开拼接处理，得到一个数据向量，以使 该数据向量用于记录该图像数据所携带的像素信息；而且本申请实施例不限定该“压平处理”的实施方式，可以采用现有的或者未来出现的任一种压平处理方法进行实施。
上述“第n个压平数据”是指一个用于表示待识别图像携带的图像信息的数据向量；而且本申请实施例不限定该“第n个压平数据”，例如，当上述“待使用特征图中第n个通道的图像数据”为64×64维的数据矩阵时，该“第n个压平数据”可以是4096维的数据向量。
S22：将第n个压平数据按照预设提取参数进行数据提取处理，得到该第n个压平数据的至少一个提取数据段。其中，n为正整数，n≤N。
其中，“预设提取参数”可以预先设定；而且本申请实施例不限定该“预设提取参数”，例如，其具体可以1024×4。其中，“1024”是指每个提取数据段的数据长度；“4”是指提取数据段的个数。
上述“数据提取处理”用于针对一个数据向量进行数据划分操作；而且本申请实施例不限定该“数据提取处理”的实施方式，例如，可以采用现有的或者未来出现的任一种能够将一个数据向量划分成多个提取数据段的方法(例如，尺寸调整(resize)方法)进行实施。
上述“第n个压平数据的至少一个提取数据段”用于表示该第n个压平数据携带的图像信息；而且本申请实施例不限定该“第n个压平数据的至少一个提取数据段”，例如，当第n个压平数据为4096维的数据向量，且预设提取参数为1024×4时，因该第n个压平数据能够被划分成4个1024维的提取数据段，故该“第n个压平数据的至少一个提取数据段”可以包括4个1024维的提取数据段。
S23：根据N个压平数据的至少一个提取数据段，确定至少一个待使用特征。
作为示例，当N为32，且上述“第n个压平数据的至少一个提取数据段”包括4个1024维的提取数据段时，上述“N个压平数据的至少一个提取数据段”可以包括96个1024维的提取数据段，故S23具体可以包括：将第1个1024维的提取数据段、第2个1024维的提取数据段、……、以及第96个1024维的提取数据段均确定为待使用特征，以得到96个待使用特征。
基于上述S21至S23的相关内容可知，在获取到待识别图像的视觉特征之后，可以将该视觉特征拆分成多个特征数据，以便后续能够借助针对该多个特征数据分别进行编码处理的方式，确定该视觉特征的编码特征，如此有利于提高后续编码效率。
S3：分别对各个待使用特征进行特征编码处理，得到各个待使用特征的特征编码结果。
其中，“特征编码处理”可以预先设定；而且本申请实施例不限定该“特征编码处理”的实施方式，例如，可以采用现有的或者未来出现的任一种编码方法进行实施。又如，可以借助预先构建的特征编码模型进行实施。
上述“特征编码模型”用于针对该特征编码模型的输入数据进行编码处理；而且本申请实施例不限定该“特征编码模型”的模型结构，例如，其可以包括5个编码网络。需要说明的是，本申请实施例不限定“编码网络”的实施方式，例如，可以采用任一种编码方法(例如，transformer模型中Encoder模块，或者，conformer结构)进行实施。
另外，本申请实施例不限定“特征编码模型”的构建方法，例如，可以采用现有的或者未来出现的任一种模型构建方法进行实施。又如，还可以采用下文 方法实施例三所示的模型构建方法进行实施。
第k个待使用特征的特征编码结果用于表示该第k个待使用特征所携带的图像信息(例如，摩尔纹信息)；而且本申请实施例不限定该“第k个待使用特征的特征编码结果”的确定过程，例如，其具体可以为：将第k个待使用特征输入预先构建的特征编码模型，得到该特征编码模型输出的该第k个待使用特征的特征编码结果。其中，k为正整数，k≤K，K为正整数，K表示待使用特征的个数(例如，96)。
S4：根据至少一个待使用特征的特征编码结果，确定待识别图像是否为翻拍图像。
本申请实施例不限定S4的实施方式，例如，其具体可以为：将至少一个待使用特征的特征编码结果输入预先训练好的翻拍图像检测模型，得到该翻拍图像检测模型输出的检测结果，以使该检测结果用于表示该待识别图像是否为翻拍图像。
上述“翻拍图像检测模型”用于针对该翻拍图像检测模型的输入数据进行翻拍图像检测识别处理；而且本申请实施例不限定该“翻拍图像检测模型”，例如，其可以是现有的或者未来出现的任一种机器学习模型。需要说明的是，本申请实施例也不限定该“翻拍图像检测模型”的训练过程，例如，可以采用现有的或者未来出现的任一种模型训练方法进行实施。
另外，为了提高翻拍图像的识别效果，本申请实施例还提供了S4的另一种可能的实施方式，其具体可以包括S41-S43：
S41：将至少一个待使用特征的特征编码结果进行拼接，得到待识别图像的编码特征。
其中，“待识别图像的编码特征”用于表示该待识别图像携带的图像信息(尤其是，摩尔纹信息等)。
另外，本申请实施例不限定上述“拼接”的实施方式，可以采用现有的或者未来出现的任一种拼接方法进行实施。
S42：对待识别图像的编码特征进行翻拍图像识别处理，得到待识别图像的识别结果。
其中，“翻拍图像识别处理”用于识别一个图像数据的编码特征是否属于翻拍图像所具有的特征；而且本申请实施例不限定该“翻拍图像识别处理”，例如，其可以借助预先构建的翻拍识别模型进行实施。
上述“翻拍识别模型”用于针对该翻拍识别模型的输入数据进行翻拍图像识别处理；而且本申请实施例不限定该“翻拍识别模型”的模型结构，例如，其可以包括线性处理层和全连接层，且该全连接层的输入数据包括该线性处理层的输出数据。
另外，本申请实施例不限定上述“线性处理层”的实施方式，例如，其可以采用现有的或者未来出现的任一种线性处理方法(例如，transformer模型中linear模块)进行实施。同样，本申请实施例也不限定上述“全连接层”的实施方式，例如，其可以采用现有的或者未来出现的任一种全连接处理方法(例如，transformer模型中softmax模块)进行实施。
此外，本申请实施例不限定“翻拍识别模型”的构建方法，例如，可以采用现有的或者未来出现的任一种模型构建方法进行实施。又如，还可以采用下文 方法实施例三所示的 模型构建方法进行实施。
上述“待识别图像的识别结果”用于表示该待识别图像属于翻拍图像的可能性；而且本申请实施例不限定该“待识别图像的识别结果”的表示方式，例如，其可以包括第一概率和第二概率。其中，“第一概率”用于表示待识别图像属于翻拍图像的可能性；“第二概率”用于表示待识别图像不属于翻拍图像的可能性。
另外，本申请实施例不限定S42的实施方式，例如，其具体可以包括：将待识别图像的编码特征输入预先构建的翻拍识别模型，得到该翻拍识别模型输出的该待识别图像的识别结果。
S43：根据待识别图像的识别结果，确定待识别图像是否为翻拍图像。
本申请实施例中，在获取到待识别图像的识别结果之后，可以判断该待识别图像的识别结果是否达到预设翻拍条件；若达到，则可以确定该待识别图像为翻拍图像；若不达到，则可以确定该待识别图像不是翻拍图像。其中，“预设翻拍条件”可以预先设定，例如，该“预设翻拍条件”具体可以为：第一概率高于第一阈值，和/或，第二概率低于第二阈值。
基于上述S1至S4的相关内容可知，对于本申请实施例提供的翻拍图像识别方法来说，在获取到待识别图像之后，先对该待识别图像进行视觉特征提取处理，得到该待识别图像的视觉特征；再对该视觉特征进行特征拆分处理，得到至少一个待使用特征；然后，分别对各个该待使用特征进行特征编码处理，得到各个该待使用特征的特征编码结果；最后，根据该至少一个待使用特征的特征编码结果，确定该待识别图像是否为翻拍图像。
可见，因上述“至少一个待使用特征的特征编码结果”能够准确地表示出待识别图像携带的图像信息，使得该“至少一个待使用特征的特征编码结果”能够准确地表示出待识别图像携带的摩尔纹信息，从而使得该“至少一个待使用特征的特征编码结果”能够准确地表示出该待识别图像是否为翻拍图像，如此使得后续能够基于该“至少一个待使用特征的特征编码结果”，识别出一个图像数据是否为翻拍图像。
方法实施例二
为了提高翻拍图像的识别效果，本申请实施例还提供了翻拍图像识别方法的另一种可能的实施方式，在该实施方式中，该翻拍图像识别方法除了包括上述S1、S2、以及S4以外，可以还包括S5-S7：
S5：确定各个待使用特征的加权权重值。
其中，第k个待使用特征的加权权重值用于表示该第k个待使用特征所表征的图像信息的重要程度；而且本申请实施例不限定该“第k个待使用特征的加权权重值”的获取方式，例如，可以预先设定。其中，k为正整数，k≤K，K为正整数，K表示待使用特征的个数(例如，96)。
另外，为了提高翻拍图像的识别灵活性，本申请实施例还提供了S5的另一种可能的实施方式，其具体可以包括：将各个待使用特征输入预先构建的权重确定模型，得到该权重确定模型输出的各个待使用特征的加权权重值。
上述“权重确定模型”用于针对该权重确定模型的输入数据进行信息重要度衡量处理；而且本申请实施例不限定该“权重确定模型”，例如，其可以是任一种机器学习模型(例如， 1×1的卷积网络)即可。
另外，本申请实施例不限定上述“权重确定模型”的构建过程，例如，其可以采用现有的或者未来出现的任一种模型构建方法进行实施。又如，还可以采用下文 方法实施例三所示的模型构建方法进行实施。
S6：将各个待使用特征分别与各个待使用特征的加权权重值进行相乘，得到各个加权特征。
作为示例，S6具体可以包括：将第k个待使用特征与该第k个待使用特征的加权权重值进行相乘，得到第k个加权特征，以使该第k个加权特征能够更准确地表示出该第k个待使用特征所携带的图像信息及其对后续图像翻拍识别处理所造成的影响程度。
S7：分别对各个加权特征进行特征编码处理，得到各个待使用特征的特征编码结果。
需要说明的是，S7的相关内容类似于上文S3的相关内容，只需将上文S3的相关内容中“待使用特征”替换为“加权特征”即可。
基于上述S5至S7的相关内容可知，对于待识别图像来说，可以参考该待识别图像的至少一个待使用特征及其加权权重值，对该待识别图像的至少一个待使用特征进行编码处理，得到各个待使用特征的特征编码结果，以使各个待使用特征的特征编码结果能够更好的表示出该待识别图像携带的图像信息(尤其是，摩尔纹信息)，如此有利于提高翻拍图像的识别准确性。
方法实施例三
为了提高上述模型(例如，视觉特征提取模型、特征编码模型、权重确定模型、翻拍识别模型等)的构建效果，本申请实施例还提供了一种模型构建方法，其具体可以包括步骤11-步骤14：
步骤11：获取样本图像和该样本图像的实际标签。
其中，上述“样本图像”是指在模型构建过程中所需使用的图像数据；而且本申请实施例不限定“样本图像”的个数。
上述“样本图像的实际标签”用于表示该样本图像实际是否为翻拍图像；而且本申请实施例不限定该“样本图像的实际标签”的获取过程，例如，可以通过人工标注的方式进行实施。
另外，本申请实施例不限定上述“样本图像的实际标签”的表示方式，例如，其可以包括第三概率和第四概率。其中，“第三概率”用于表示样本图像实际属于翻拍图像的可能性；“第四概率”用于表示样本图像实际不属于翻拍图像的可能性。
步骤12：根据样本图像，确定至少一个待使用图像。
其中，“待使用图像”用于表示样本图像中部分图像信息或者全部图像信息。
另外，本申请实施例不限定上述“至少一个待使用图像”，例如，其可以包括一个第一图像样本、H个第二图像样本、以及G个第三图像样本。其中，H为正整数，G为正整数。需要说明的是，本申请实施例不限定H与G之间的关系，例如，G＝2×H。
上述“第一图像样本”用于记录样本图像中全部图像信息；而且本申请实施例不限定该“第一图像样本”的确定过程，例如，可以直接将该样本图像确定为第一图像样本。
上述“第二图像样本”用于记录样本图像中第一裁剪比例下(例如，70％)的图像信息；而且本申请实施例不限定上述“H个第二图像样本”的确定过程，例如，可以按照第一裁剪比例，对该样本图像进行第h次随机裁剪处理，得到第h个第二图像样本，以使该第h个第二图像样本携带的图像信息与该样本图像携带的图像信息之间的相同比例达到上述“第一裁剪比例”。
上述“第三图像样本”用于记录样本图像中第二裁剪比例下(例如，40％)的图像信息；而且本申请实施例不限定上述“G个第三图像样本”的确定过程，例如，可以按照第二裁剪比例，对该样本图像进行第g次随机裁剪处理，得到第g个第三图像样本，以使该第g个第三图像样本携带的图像信息与该样本图像携带的图像信息之间的相同比例达到上述“第二裁剪比例”。
此外，本申请实施例不限定步骤12的实施方式，例如，其具体可以包括步骤121-步骤122：
步骤121：按照预设规则，对样本图像进行裁剪处理，得到至少一个裁剪图像。
其中，预设规则可以预先设定，例如，其具体可以包括：按照第一裁剪比例对样本图像进行H次裁剪处理，并按照第二裁剪比例对样本图像进行G次裁剪处理。
上述“裁剪图像”用于表示样本图像中部分图像信息。
步骤122：根据样本图像和至少一个裁剪图像，确定至少一个待使用图像。
本申请实施例不限定步骤122的实施方式，例如，其具体可以为：直接将样本图像、以及各个裁剪图像均确定为待使用图像。又如，步骤122具体可以为：先按照样本图像的图像尺寸，对各个裁剪图像进行数据填充处理，得到各个填充图像，以使各个填充图像的图像尺寸与该样本图像的图像尺寸保持一致；再将样本图像、以及各个填充图像均确定为待使用图像。
基于上述步骤12的相关内容可知，在获取到样本图像之后，可以参考该样本图像中不同区域的图像信息，确定不同待使用图像，以使不同待使用图像能够表示该样本图像中不同区域所携带的图像信息，从而使得基于这些待使用图像构建的各个模型具有较好的图像区域特征提取性能，如此有利于提高特征提取效果，从而有利于提高翻拍图像的识别效果。
步骤13：将样本图像的实际标签确定为各个待使用图像的实际标签。
本申请实施例中，在获取到各个待使用图像的之后，可以依据样本图像的实际标签，确定各个待使用图像的实际标签，以使各个待使用图像的实际标签与该样本图像的实际标签保持一致。
步骤14：利用至少一个待使用图像和该至少一个待使用图像的实际标签，构建视觉特征提取模型、特征编码模型、权重确定模型、以及翻拍识别模型中的至少一个。
作为示例，步骤14具体可以包括步骤141-步骤144：
步骤141：将各个待使用图像输入待训练模型，得到该待训练模型输出的各个待使用图像的识别结果。
其中，“待训练模型”用于针对该待训练模型的输入数据进行翻拍图像识别处理；而且本申请实施例不限定该“待训练模型”，例如，其可以是一种机器学习模型。
另外，本申请实施例不限定上述“待训练模型”的模型结构，例如，如图2所示，“待训练模型”具体可以包括视觉特征提取层、特征拆分层、权重确定层、特征加权层、特征编码层、特征拼接层和翻拍识别层。其中，翻拍识别层的输入数据包括特征拼接层的输出数据；特征拼接层的输入数据包括特征编码层的输出数据；特征编码层的输入数据包括特征加权层的输出数据；特征加权层的输入数据包括权重确定层的输出数据和特征拆分层的输出数据；权重确定层的输入数据包括特征拆分层的输出数据；特征拆分层的输入数据包括视觉特征提取层的输出数据。
上述“视觉特征提取层”用于针对一个图像数据进行视觉特征提取处理；而且本申请实施例不限定该“视觉特征提取层”的网络结构，例如，其可以采用上文“视觉特征提取模型”的模型结构进行实施。
上述“特征拆分层”用于针对该特征拆分层的输入数据进行特征拆分处理；而且本申请实施例不限定该“特征拆分层”的工作原理，例如，其可以采用上文S21-S23所示的特征拆分处理过程进行实施。
上述“权重确定层”用于针对该权重确定层的输入数据进行信息重要度衡量处理；而且本申请实施例不限定该“权重确定层”的网络结构，例如，其可以采用上文“权重确定模型”的模型结构进行实施。
上述“特征加权层”用于针对该权重确定层的输入数据进行加权处理；而且本申请实施例不限定该“特征加权层”的工作原理，例如，其可以采用上文S6所示的加权处理过程进行实施。
上述“特征编码层”用于针对该特征编码层的输入数据进行编码处理；而且本申请实施例不限定该“特征编码层”的网络结构，例如，其可以采用上文“特征编码模型”的模型结构进行实施。
上述“特征拼接层”用于针对该特征拼接层的输入数据进行拼接处理；而且本申请实施例不限定该“特征拼接层”的工作原理，可以采用现有的或者未来出现的任一种拼接方法进行实施。
上述“翻拍识别层”用于针对该翻拍识别层的输入数据进行翻拍数据识别处理；而且本申请实施例不限定该“翻拍识别层”的网络结构，例如，其可以采用上文“翻拍识别模型”的模型结构进行实施。
上述“待使用图像的识别结果”用于表示该待使用图像预测属于翻拍图像的可能性。
步骤142：判断是否达到预设停止条件，若是，则执行步骤144；若否，则执行步骤143。
其中，“预设停止条件”可以预先设定；而且本申请实施例不限定该“预设停止条件”，例如，其具体可以为待训练模型的模型损失值低于预设损失值阈值；也可以为该待训练模型的模型损失值的变化率低于预设变化率阈值；还可以为该待训练模型的更新次数达到预设次数阈值。
上述“待训练模型的模型损失值”用于表示该待训练模型的翻拍图像识别性能；而且本申请实施例不限定该“待训练模型的模型损失值”的确定过程，可以采用现有的或者未 来出现的任一种模型损失值计算方法进行实施。
步骤143：根据至少一个待使用图像的识别结果和该至少一个待使用图像的实际标签，更新待训练模型，并返回执行步骤141。
本申请实施例中，在确定当前轮的待训练模型没有达到预设停止条件之后，可以确定该待训练模型的翻拍图像识别性能仍然比较差，故可以依据至少一个待使用图像的识别结果和该至少一个待使用图像的实际标签之间的差异性，更新该待训练模型，以使更新后的待训练模型具有更好的翻拍图像识别性能，并基于更新后的待训练模型继续执行步骤141及其后续步骤，以实现针对该待训练模型的下一轮训练过程。
步骤144：根据待训练模型，确定视觉特征提取模型、特征编码模型、权重确定模型、以及翻拍识别模型中的至少一个。
本申请实施例中，在确定当前轮的待训练模型达到预设停止条件之后，可以确定该待训练模型具有较好的翻拍图像识别性能，故可以直接根据该待训练模型，确定视觉特征提取模型、特征编码模型、权重确定模型、以及翻拍识别模型中的至少一个；而且该确定过程具体可以为：根据该待训练模型中视觉特征提取层，确定视觉特征提取模型(例如，直接将该待训练模型中视觉特征提取层，确定为视觉特征提取模型)；根据该待训练模型中特征编码层，确定特征编码模型(例如，直接将该待训练模型中特征编码层，确定为特征编码模型)；根据该待训练模型中权重确定层，确定权重确定模型(例如，直接将该待训练模型中权重确定层，确定为权重确定模型)；根据该待训练模型中翻拍识别层，确定翻拍识别模型(例如，直接将该待训练模型中翻拍识别层，确定为翻拍识别模型)。
基于上述步骤11至步骤14的相关内容可知，在获取到样本图像及其实际标签之后，可以先参考该样本图像及其实际标签，确定至少一个待使用图像及其实际标签；再利用至少一个待使用图像及其实际标签，构建视觉特征提取模型、特征编码模型、权重确定模型、以及翻拍识别模型中的至少一个，以使构建好的各个模型均具有较好的模型性能，如此有利于提高翻拍图像的识别性能。
基于上述方法实施例提供的翻拍图像识别方法，本申请实施例还提供了一种翻拍图像识别装置，下面结合附图进行解释和说明。
装置实施例
装置实施例提供的翻拍图像识别装置的技术详情，请参照上述方法实施例。
参见图3，该图为本申请实施例提供的一种翻拍图像识别装置的结构示意图。
本申请实施例提供的翻拍图像识别装置300，包括：
提取单元301，用于在获取到待识别图像之后，对所述待识别图像进行视觉特征提取处理，得到所述待识别图像的视觉特征；
拆分单元302，用于对所述视觉特征进行特征拆分处理，得到至少一个待使用特征；
编码单元303，用于分别对各个所述待使用特征进行特征编码处理，得到各个所述待使用特征的特征编码结果；
识别单元304，用于根据所述至少一个待使用特征的特征编码结果，确定所述待识别图像是否为翻拍图像。
在一种可能的实施方式中，所述视觉特征包括待使用特征图，且所述待使用特征图包括N个通道的图像数据；其中，N为正整数；
所述拆分单元302，具体用于：对所述待使用特征图中第n个通道的图像数据进行压平处理，得到第n个压平数据；其中，n为正整数，n≤N；将所述第n个压平数据按照预设提取参数进行数据提取处理，得到所述第n个压平数据的至少一个提取数据段；其中，n为正整数，n≤N；根据N个压平数据的至少一个提取数据段，确定所述至少一个待使用特征。
在一种可能的实施方式中，所述翻拍图像识别装置300还包括：
确定单元，用于确定各个所述待使用特征的加权权重值；
加权单元，用于将各个所述待使用特征分别与各个所述待使用特征的加权权重值进行相乘，得到各个加权特征；
所述编码单元303，具体用于：分别对各个所述加权特征进行特征编码处理，得到各个所述待使用特征的特征编码结果。
在一种可能的实施方式中，所述确定单元，具体用于：将各个所述待使用特征输入预先构建的权重确定模型，得到所述权重确定模型输出的各个所述待使用特征的加权权重值。
在一种可能的实施方式中，所述识别单元304，包括：
拼接子单元，用于将所述至少一个待使用特征的特征编码结果进行拼接，得到所述待识别图像的编码特征；
处理子单元，用于对所述待识别图像的编码特征进行翻拍图像识别处理，得到所述待识别图像的识别结果；
确定子单元，用于根据所述待识别图像的识别结果，确定所述待识别图像是否为翻拍图像。
在一种可能的实施方式中，所述处理子单元，具体用于：将所述待识别图像的编码特征输入预先构建的翻拍识别模型，得到所述翻拍识别模型输出的所述待识别图像的识别结果。
在一种可能的实施方式中，所述翻拍图像识别装置300还包括：
构建单元，用于获取样本图像和所述样本图像的实际标签；其中，所述样本图像的实际标签用于表示所述样本图像实际是否为翻拍图像；根据所述样本图像，确定至少一个待使用图像；将所述样本图像的实际标签确定为各个待使用图像的实际标签；利用所述至少一个待使用图像和所述至少一个待使用图像的实际标签，构建所述翻拍识别模型。
在一种可能的实施方式中，所述构建单元，包括：
识别子单元，用于将各个所述待使用图像输入待训练模型，得到所述待训练模型输出的各个所述待使用图像的识别结果；
更新子单元，用于根据所述至少一个待使用图像的识别结果和所述至少一个待使用图像的实际标签，更新所述待训练模型，并返回所述识别子单元继续执行所述将各个所述待使用图像输入待训练模型的步骤；
构建子单元，用于在达到预设停止条件时，根据所述待训练模型，确定所述翻拍识别 模型。
在一种可能的实施方式中，所述待训练模型包括：视觉特征提取层、特征拆分层、权重确定层、特征加权层、特征编码层、特征拼接层和翻拍识别层；所述翻拍识别层的输入数据包括所述特征拼接层的输出数据；所述特征拼接层的输入数据包括所述特征编码层的输出数据；所述特征编码层的输入数据包括所述特征加权层的输出数据；所述特征加权层的输入数据包括所述权重确定层的输出数据和所述特征拆分层的输出数据；所述权重确定层的输入数据包括所述特征拆分层的输出数据；所述特征拆分层的输入数据包括所述视觉特征提取层的输出数据；
所述构建子单元，具体用于：根据所述待训练模型中翻拍识别层，确定所述翻拍识别模型。
在一种可能的实施方式中，所述拆分单元302，具体用于：按照预设规则，对所述样本图像进行裁剪处理，得到至少一个裁剪图像；根据所述样本图像和所述至少一个裁剪图像，确定至少一个待使用图像。
在一种可能的实施方式中，所述翻拍图像识别装置300还包括：
调整单元，用于在获取到设备采集图像之后，对所述设备采集图像按照预设尺寸进行调整处理，得到所述待识别图像。
基于上述翻拍图像识别装置300的相关内容可知，对于翻拍图像识别装置300来说，在获取到待识别图像之后，先对该待识别图像进行视觉特征提取处理，得到该待识别图像的视觉特征；再对该视觉特征进行特征拆分处理，得到至少一个待使用特征；然后，分别对各个该待使用特征进行特征编码处理，得到各个该待使用特征的特征编码结果；最后，根据该至少一个待使用特征的特征编码结果，确定该待识别图像是否为翻拍图像。其中，因上述“至少一个待使用特征的特征编码结果”能够准确地表示出待识别图像携带的图像信息，使得该“至少一个待使用特征的特征编码结果”能够准确地表示出待识别图像携带的摩尔纹信息，从而使得该“至少一个待使用特征的特征编码结果”能够准确地表示出该待识别图像是否为翻拍图像，如此使得后续能够基于该“至少一个待使用特征的特征编码结果”，识别出一个图像数据是否为翻拍图像。
进一步地，本申请实施例还提供了一种设备，所述设备包括处理器以及存储器：
所述存储器用于存储计算机程序；
所述处理器用于根据所述计算机程序执行本申请实施例提供的翻拍图像识别方法的任一实施方式。
进一步地，本申请实施例还提供了一种计算机可读存储介质，所述计算机可读存储介质用于存储计算机程序，所述计算机程序用于执行本申请实施例提供的翻拍图像识别方法的任一实施方式。
进一步地，本申请实施例还提供了一种计算机程序产品，所述计算机程序产品在终端设备上运行时，使得所述终端设备执行本申请实施例提供的翻拍图像识别方法的任一实施方式。
应当理解，在本申请中，“至少一个(项)”是指一个或者多个，“多个”是指两个或两 个以上。“和/或”，用于描述关联对象的关联关系，表示可以存在三种关系，例如，“A和/或B”可以表示：只存在A，只存在B以及同时存在A和B三种情况，其中A，B可以是单数或者复数。字符“/”一般表示前后关联对象是一种“或”的关系。“以下至少一项(个)”或其类似表达，是指这些项中的任意组合，包括单项(个)或复数项(个)的任意组合。例如，a，b或c中的至少一项(个)，可以表示：a，b，c，“a和b”，“a和c”，“b和c”，或“a和b和c”，其中a，b，c可以是单个，也可以是多个。
以上所述，仅是本发明的较佳实施例而已，并非对本发明作任何形式上的限制。虽然本发明已以较佳实施例揭露如上，然而并非用以限定本发明。任何熟悉本领域的技术人员，在不脱离本发明技术方案范围情况下，都可利用上述揭示的方法和技术内容对本发明技术方案做出许多可能的变动和修饰，或修改为等同变化的等效实施例。因此，凡是未脱离本发明技术方案的内容，依据本发明的技术实质对以上实施例所做的任何简单修改、等同变化及修饰，均仍属于本发明技术方案保护的范围内。

Claims (15)

  1. 一种翻拍图像识别方法，其特征在于，所述方法包括：
    在获取到待识别图像之后，对所述待识别图像进行视觉特征提取处理，得到所述待识别图像的视觉特征；
    对所述视觉特征进行特征拆分处理，得到至少一个待使用特征；
    分别对各个所述待使用特征进行特征编码处理，得到各个所述待使用特征的特征编码结果；
    根据所述至少一个待使用特征的特征编码结果，确定所述待识别图像是否为翻拍图像。
  2. 根据权利要求1所述的方法，其特征在于，所述视觉特征包括待使用特征图，且所述待使用特征图包括N个通道的图像数据；其中，N为正整数；
    所述至少一个待使用特征的确定过程，包括：
    对所述待使用特征图中第n个通道的图像数据进行压平处理，得到第n个压平数据；其中，n为正整数，n≤N；
    将所述第n个压平数据按照预设提取参数进行数据提取处理，得到所述第n个压平数据的至少一个提取数据段；其中，n为正整数，n≤N；
    根据N个压平数据的至少一个提取数据段，确定所述至少一个待使用特征。
  3. 根据权利要求1所述的方法，其特征在于，所述方法还包括：
    确定各个所述待使用特征的加权权重值；
    将各个所述待使用特征分别与各个所述待使用特征的加权权重值进行相乘，得到各个加权特征；
    所述分别对各个所述待使用特征进行特征编码处理，得到各个所述待使用特征的特征编码结果，包括：
    分别对各个所述加权特征进行特征编码处理，得到各个所述待使用特征的特征编码结果。
  4. 根据权利要求3所述的方法，其特征在于，所述确定各个所述待使用特征的加权权重值，包括：
    将各个所述待使用特征输入预先构建的权重确定模型，得到所述权重确定模型输出的各个所述待使用特征的加权权重值。
  5. 根据权利要求1所述的方法，其特征在于，所述根据所述至少一个待使用特征的特征编码结果，确定所述待识别图像是否为翻拍图像，包括：
    将所述至少一个待使用特征的特征编码结果进行拼接，得到所述待识别图像的编码特征；
    对所述待识别图像的编码特征进行翻拍图像识别处理，得到所述待识别图像的识别结果；
    根据所述待识别图像的识别结果，确定所述待识别图像是否为翻拍图像。
  6. 根据权利要求5所述的方法，其特征在于，所述对所述待识别图像的编码特征进行翻拍图像识别处理，得到所述待识别图像的识别结果，包括：
    将所述待识别图像的编码特征输入预先构建的翻拍识别模型，得到所述翻拍识别模型输出的所述待识别图像的识别结果。
  7. 根据权利要求6所述的方法，其特征在于，所述翻拍识别模型的构建过程，包括：
    获取样本图像和所述样本图像的实际标签；其中，所述样本图像的实际标签用于表示所述样本图像实际是否为翻拍图像；
    根据所述样本图像，确定至少一个待使用图像；
    将所述样本图像的实际标签确定为各个待使用图像的实际标签；
    利用所述至少一个待使用图像和所述至少一个待使用图像的实际标签，构建所述翻拍识别模型。
  8. 根据权利要求7所述的方法，其特征在于，所述利用所述至少一个待使用图像和所述至少一个待使用图像的实际标签，构建所述翻拍识别模型，包括：
    将各个所述待使用图像输入待训练模型，得到所述待训练模型输出的各个所述待使用图像的识别结果；
    根据所述至少一个待使用图像的识别结果和所述至少一个待使用图像的实际标签，更新所述待训练模型，并继续执行所述将各个所述待使用图像输入待训练模型的步骤，直至在达到预设停止条件时，根据所述待训练模型，确定所述翻拍识别模型。
  9. 根据权利要求8所述的方法，其特征在于，所述待训练模型包括：视觉特征提取层、特征拆分层、权重确定层、特征加权层、特征编码层、特征拼接层和翻拍识别层；所述翻拍识别层的输入数据包括所述特征拼接层的输出数据；所述特征拼接层的输入数据包括所述特征编码层的输出数据；所述特征编码层的输入数据包括所述特征加权层的输出数据；所述特征加权层的输入数据包括所述权重确定层的输出数据和所述特征拆分层的输出数据；所述权重确定层的输入数据包括所述特征拆分层的输出数据；所述特征拆分层的输入数据包括所述视觉特征提取层的输出数据；
    所述根据所述待训练模型，确定所述翻拍识别模型，包括：
    根据所述待训练模型中翻拍识别层，确定所述翻拍识别模型。
  10. 根据权利要求7所述的方法，其特征在于，所述根据所述样本图像，确定至少一个待使用图像，包括：
    按照预设规则，对所述样本图像进行裁剪处理，得到至少一个裁剪图像；
    根据所述样本图像和所述至少一个裁剪图像，确定至少一个待使用图像。
  11. 根据权利要求1所述的方法，其特征在于，所述待识别图像的获取过程，包括：
    在获取到设备采集图像之后，对所述设备采集图像按照预设尺寸进行调整处理，得到所述待识别图像。
  12. 一种翻拍图像识别装置，其特征在于，包括：
    提取单元，用于在获取到待识别图像之后，对所述待识别图像进行视觉特征提取处理，得到所述待识别图像的视觉特征；
    拆分单元，用于对所述视觉特征进行特征拆分处理，得到至少一个待使用特征；
    编码单元，用于分别对各个所述待使用特征进行特征编码处理，得到各个所述待使用 特征的特征编码结果；
    识别单元，用于根据所述至少一个待使用特征的特征编码结果，确定所述待识别图像是否为翻拍图像。
  13. 一种设备，其特征在于，所述设备包括处理器以及存储器：
    所述存储器用于存储计算机程序；
    所述处理器用于根据所述计算机程序执行权利要求1-11中任一项所述的方法。
  14. 一种计算机可读存储介质，其特征在于，所述计算机可读存储介质用于存储计算机程序，所述计算机程序用于执行权利要求1-11中任一项所述的方法。
  15. 一种计算机程序产品，其特征在于，所述计算机程序产品在终端设备上运行时，使得所述终端设备执行权利要求1-11中任一项所述的方法。
PCT/CN2022/119635 2021-10-29 2022-09-19 一种翻拍图像识别方法及其相关设备 Ceased WO2023071609A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202111275642.9A CN114005019B (zh) 2021-10-29 2021-10-29 一种翻拍图像识别方法及其相关设备
CN202111275642.9 2021-10-29

Publications (1)

Publication Number Publication Date
WO2023071609A1 true WO2023071609A1 (zh) 2023-05-04

Family

ID=79925649

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2022/119635 Ceased WO2023071609A1 (zh) 2021-10-29 2022-09-19 一种翻拍图像识别方法及其相关设备

Country Status (2)

Country Link
CN (1) CN114005019B (zh)
WO (1) WO2023071609A1 (zh)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116958795A (zh) * 2023-06-30 2023-10-27 北京房多多信息技术有限公司 翻拍图像的识别方法、装置、电子设备及存储介质

Families Citing this family (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114005019B (zh) * 2021-10-29 2023-09-22 北京有竹居网络技术有限公司 一种翻拍图像识别方法及其相关设备
CN116205250A (zh) * 2022-11-11 2023-06-02 中国银联股份有限公司 一种二维码的唯一化标记的验证方法及其验证系统
CN116310571A (zh) * 2023-03-23 2023-06-23 度小满科技(北京)有限公司 一种翻拍图像的检测方法和装置、电子设备

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111368944A (zh) * 2020-05-27 2020-07-03 支付宝(杭州)信息技术有限公司 翻拍图像、证件照识别、模型训练方法、装置及电子设备
CN111476269A (zh) * 2020-03-04 2020-07-31 中国平安人寿保险股份有限公司 均衡样本集构建、翻拍图像识别方法、装置、设备及介质
CN112861950A (zh) * 2021-01-29 2021-05-28 深圳前海微众银行股份有限公司 翻拍图像的识别方法、装置、电子设备及存储介质
US20210232848A1 (en) * 2018-08-30 2021-07-29 Nokia Technologies Oy Apparatus and method for processing image data
CN113344000A (zh) * 2021-06-29 2021-09-03 南京星云数字技术有限公司 证件翻拍识别方法、装置、计算机设备和存储介质
CN114005019A (zh) * 2021-10-29 2022-02-01 北京有竹居网络技术有限公司 一种翻拍图像识别方法及其相关设备

Family Cites Families (17)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP2008193409A (ja) * 2007-02-05 2008-08-21 Ricoh Co Ltd 画像処理装置
CN104598933B (zh) * 2014-11-13 2017-12-15 上海交通大学 一种基于多特征融合的图像翻拍检测方法
CN109325462B (zh) * 2018-10-11 2021-03-12 深圳斐视沃德科技有限公司 基于虹膜的人脸识别活体检测方法及装置
CN109558794B (zh) * 2018-10-17 2024-06-28 平安科技（深圳）有限公司 基于摩尔纹的图像识别方法、装置、设备和存储介质
CN109815960A (zh) * 2018-12-21 2019-05-28 深圳壹账通智能科技有限公司 基于深度学习的翻拍图像识别方法、装置、设备及介质
CN109815970B (zh) * 2018-12-21 2023-04-07 平安科技（深圳）有限公司 翻拍图像的识别方法、装置、计算机设备和存储介质
CN109886275A (zh) * 2019-01-16 2019-06-14 深圳壹账通智能科技有限公司 翻拍图像识别方法、装置、计算机设备和存储介质
CN110110715A (zh) * 2019-04-30 2019-08-09 北京金山云网络技术有限公司 文本检测模型训练方法、文本区域、内容确定方法和装置
CN110543815B (zh) * 2019-07-22 2024-03-08 平安科技（深圳）有限公司 人脸识别模型的训练方法、人脸识别方法、装置、设备及存储介质
CN110717450B (zh) * 2019-10-09 2023-02-03 深圳大学 自动识别原始文档的翻拍图像的训练方法和检测方法
CN111027584A (zh) * 2019-10-23 2020-04-17 宋飞 一种课堂行为识别方法及设备
CN111191568B (zh) * 2019-12-26 2024-06-14 中国平安人寿保险股份有限公司 翻拍图像识别方法、装置、设备及介质
CN111461143A (zh) * 2020-03-31 2020-07-28 珠海格力电器股份有限公司 一种图片翻拍识别方法和装置及电子设备
CN111462120B (zh) * 2020-06-17 2020-09-29 熵智科技(深圳)有限公司 一种基于语义分割模型缺陷检测方法、装置、介质及设备
CN112070714B (zh) * 2020-07-29 2024-02-20 西安工业大学 一种基于局部三元计数特征的翻拍图像检测方法
CN112883256B (zh) * 2021-01-11 2024-05-17 北京达佳互联信息技术有限公司 多任务处理方法、装置、电子设备及存储介质
CN112767394A (zh) * 2021-03-04 2021-05-07 重庆赛迪奇智人工智能科技有限公司 图像检测方法、装置和设备

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20210232848A1 (en) * 2018-08-30 2021-07-29 Nokia Technologies Oy Apparatus and method for processing image data
CN111476269A (zh) * 2020-03-04 2020-07-31 中国平安人寿保险股份有限公司 均衡样本集构建、翻拍图像识别方法、装置、设备及介质
CN111368944A (zh) * 2020-05-27 2020-07-03 支付宝(杭州)信息技术有限公司 翻拍图像、证件照识别、模型训练方法、装置及电子设备
CN112861950A (zh) * 2021-01-29 2021-05-28 深圳前海微众银行股份有限公司 翻拍图像的识别方法、装置、电子设备及存储介质
CN113344000A (zh) * 2021-06-29 2021-09-03 南京星云数字技术有限公司 证件翻拍识别方法、装置、计算机设备和存储介质
CN114005019A (zh) * 2021-10-29 2022-02-01 北京有竹居网络技术有限公司 一种翻拍图像识别方法及其相关设备

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116958795A (zh) * 2023-06-30 2023-10-27 北京房多多信息技术有限公司 翻拍图像的识别方法、装置、电子设备及存储介质

Also Published As

Publication number Publication date
CN114005019B (zh) 2023-09-22
CN114005019A (zh) 2022-02-01

Similar Documents

Publication Publication Date Title
WO2023071609A1 (zh) 一种翻拍图像识别方法及其相关设备
CN115050077B (zh) 情绪识别方法、装置、设备及存储介质
CN112200817B (zh) 基于图像的天空区域分割和特效处理方法、装置及设备
CN114140831B (zh) 人体姿态估计方法、装置、电子设备及存储介质
WO2021175040A1 (zh) 视频处理方法及相关装置
WO2023050650A1 (zh) 动画视频生成方法、装置、设备及存储介质
CN114021646B (zh) 一种图像描述文本确定方法及其相关设备
CN111401374A (zh) 基于多任务的模型训练方法、字符识别方法及装置
CN111046747B (zh) 人群计数模型的训练方法、人群计数方法、装置和服务器
CN118467778B (zh) 视频信息摘要生成方法、装置、电子装置和存储介质
CN113221718B (zh) 公式识别方法、装置、存储介质和电子设备
TWI817896B (zh) 機器學習方法以及裝置
CN113343958A (zh) 一种文本识别方法、装置、设备及介质
JP2023155898A5 (zh)
CN119383289A (zh) 一种视频生成方法、装置、设备及介质
CN119131348A (zh) 模型训练方法及火灾检测方法
CN114240770A (zh) 一种图像处理方法、装置、服务器及存储介质
CN112669244B (zh) 人脸图像增强方法、装置、计算机设备以及可读存储介质
CN115909408B (zh) 一种基于Transformer网络的行人重识别方法及装置
CN118433446A (zh) 视频优化处理方法、系统、设备及存储介质
CN117391201A (zh) 问答方法、装置及电子设备
CN111160262A (zh) 一种融合人体关键点检测的人像分割方法
CN115115968A (zh) 视频质量评价方法、装置及计算机可读存储介质
CN113518229B (zh) 环路滤波网络的训练方法、装置、计算机设备和存储介质
CN113610065B (zh) 笔迹识别方法和装置

Legal Events

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

Ref document number: 22885496

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 22885496

Country of ref document: EP

Kind code of ref document: A1