WO2023020176A1 - 图像识别方法和装置 - Google Patents

图像识别方法和装置 Download PDF

Info

Publication number
WO2023020176A1
WO2023020176A1 PCT/CN2022/106160 CN2022106160W WO2023020176A1 WO 2023020176 A1 WO2023020176 A1 WO 2023020176A1 CN 2022106160 W CN2022106160 W CN 2022106160W WO 2023020176 A1 WO2023020176 A1 WO 2023020176A1
Authority
WO
WIPO (PCT)
Prior art keywords
recognition
recognition result
image
card
recognized
Prior art date
Application number
PCT/CN2022/106160
Other languages
English (en)
French (fr)
Inventor
朱雄威
孙逸鹏
魏翔
姚锟
韩钧宇
丁二锐
刘经拓
Original Assignee
北京百度网讯科技有限公司
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 北京百度网讯科技有限公司 filed Critical 北京百度网讯科技有限公司
Publication of WO2023020176A1 publication Critical patent/WO2023020176A1/zh

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches

Definitions

  • the present disclosure relates to the technical field of artificial intelligence, specifically to the technical field of computer vision and deep learning, specifically applicable to optical character recognition (Optical Character Recognition, OCR) and other scenarios, and in particular to an image recognition method and device.
  • OCR optical Character Recognition
  • Embodiments of the present disclosure provide an image recognition method, device, equipment, and storage medium.
  • an embodiment of the present disclosure provides an image recognition method, the method includes: acquiring an image to be recognized, which includes at least two card images; inputting the image to be recognized into a preset image recognition model to obtain A first recognition result corresponding to each card image in at least two card images, the first recognition result is used to indicate the category of the card image; according to the category indicated by the first recognition result, corresponding to the first recognition result The corresponding recognition operation is performed on the card image to obtain the second recognition result; the second recognition result is collected and output.
  • an embodiment of the present disclosure provides an image recognition device, which includes: an acquisition module configured to acquire an image to be recognized, where the image to be recognized includes at least two card images; an input module configured to The image to be recognized is input into a preset image recognition model to obtain a first recognition result corresponding to each card image in at least two card images, and the first recognition result is used to indicate the category of the card image; the recognition module is configured performing a corresponding recognition operation on the card image corresponding to the first recognition result according to the category indicated by the first recognition result to obtain a second recognition result; the output module is configured to summarize and output the second recognition result.
  • an embodiment of the present disclosure provides an electronic device, the electronic device includes one or more processors; a storage device, on which one or more programs are stored, when one or more programs are processors, so that one or more processors implement the image recognition method according to any embodiment of the first aspect.
  • an embodiment of the present disclosure provides a computer-readable medium on which a computer program is stored, and when the program is executed by a processor, the image recognition method according to any embodiment of the first aspect is implemented.
  • an embodiment of the present disclosure provides a computer program product, including a computer program.
  • the computer program is executed by a processor, the image recognition method according to any embodiment of the first aspect is implemented.
  • FIG. 1 is an exemplary system architecture diagram in which the present disclosure can be applied
  • FIG. 2 is a flowchart of an embodiment of an image recognition method according to the present disclosure
  • FIG. 3 is a schematic diagram of an application scenario of an image recognition method according to the present disclosure.
  • FIG. 4 is a flowchart of another embodiment of an image recognition method according to the present disclosure.
  • Fig. 5 is a schematic diagram of an embodiment of an image recognition device according to the present disclosure.
  • FIG. 6 is a schematic structural diagram of a computer system suitable for implementing the electronic device of the embodiment of the present disclosure.
  • FIG. 1 shows an exemplary system architecture 100 to which embodiments of the image recognition method of the present disclosure may be applied.
  • a system architecture 100 may include terminal devices 101 , 102 , 103 , a network 104 and a server 105 .
  • the network 104 is used as a medium for providing communication links between the terminal devices 101 , 102 , 103 and the server 105 .
  • Network 104 may include various connection types, such as wires, wireless communication links, or fiber optic cables, among others.
  • terminal devices 101 , 102 , 103 Users can use terminal devices 101 , 102 , 103 to interact with server 105 via network 104 to receive or send messages and the like.
  • Various communication client applications may be installed on the terminal devices 101, 102, 103, for example, image recognition applications, communication applications, and the like.
  • the terminal devices 101, 102, and 103 may be hardware or software.
  • the terminal devices 101, 102, 103 When the terminal devices 101, 102, 103 are hardware, they may be various electronic devices with display screens, including but not limited to mobile phones and notebook computers.
  • the terminal devices 101, 102, 103 When the terminal devices 101, 102, 103 are software, they can be installed in the electronic devices listed above. It can be implemented as a plurality of software or software modules (for example, for providing image recognition services), or as a single software or software module. No specific limitation is made here.
  • the server 105 may be a server that provides various services, for example, acquiring an image to be recognized; inputting the image to be recognized into a preset image recognition model to obtain a first recognition result corresponding to each card image in at least two card images ; According to the category indicated by the first recognition result, perform a corresponding recognition operation on the card image corresponding to the first recognition result to obtain a second recognition result; collect and output the second recognition results.
  • the server 105 may be hardware or software.
  • the server 105 can be implemented as a distributed server cluster composed of multiple servers, or as a single server.
  • the server is software, it can be implemented as multiple software or software modules (for example, for providing image recognition services), or as a single software or software module. No specific limitation is made here.
  • the image recognition method provided by the embodiments of the present disclosure may be executed by the server 105, or executed by the terminal devices 101, 102, 103, or executed by the server 105 and the terminal devices 101, 102, 103 in cooperation with each other.
  • each part (such as each unit, subunit, module, and submodule) included in the image recognition device may be all set in the server 105, or all in the terminal equipment 101, 102, 103, or in the In the server 105 and the terminal devices 101, 102, 103.
  • terminal devices, networks and servers in Fig. 1 are only illustrative. According to the implementation needs, there can be any number of terminal devices, networks and servers.
  • FIG. 2 shows a schematic flowchart 200 of an embodiment of an image recognition method.
  • the image recognition method includes the following steps:
  • Step 201 acquire an image to be recognized.
  • the execution subject for example, the server 105 or the terminal equipment 101, 102, 103 in FIG. 1
  • the execution subject can use a wired or wireless method from the local, such as the image acquisition device storing the image to be recognized, or storing the image to be recognized
  • the remote device acquires the image to be recognized.
  • the wireless connection method may include but not limited to 3G/4G connection, WiFi connection, Bluetooth connection, WiMAX connection, Zigbee connection, UWB (ultra wideband) connection, and other currently known or future wireless connection methods.
  • the image to be recognized includes at least two card images.
  • the card image may be an image of any card, for example, a driver's license, a driving license, a student ID card, a passport, a social security card, etc., which is not limited in the present disclosure.
  • the image to be recognized may be an image set including multiple card images, or an image mixed with at least two card images, which is not limited in the present disclosure.
  • the image to be recognized is an image
  • the image includes four card images, for example, the images of the main page of the driver's license, the auxiliary page of the driver's license, the main page of the driving license, and the auxiliary page of the driving license.
  • Step 202 Input the image to be recognized into a preset image recognition model to obtain a first recognition result corresponding to each of the at least two card images.
  • the execution subject inputs the image to be recognized into a preset image recognition model, and obtains a first recognition result corresponding to each card image in at least two card images, and the second A recognition result is used to indicate the category of the card image.
  • the category of the card image may include type information and attribute information of the card image.
  • the type information of the card image is used to indicate the type of the card image, for example, driver's license, driving license, student ID card, etc.
  • the attribute information of the card image is used to indicate the page category of the card image, for example, the card home page , card sub-page, etc.
  • the image recognition model can be trained based on sample images marked with category labels of card images.
  • the image recognition model may be trained based on artificial neural networks in existing technologies or future technologies, such as convolutional neural networks, recurrent neural networks, etc., which is not limited in the present disclosure.
  • the image to be recognized includes four card images, for example, a driver's license homepage image, a driver's license supplementary page image, a driving license home page image, and a driving license supplementary page image.
  • the first recognition result of each image in .
  • the image recognition model is obtained based on a convolutional neural network and a feature pyramid network (Feature Pyramid Networks, FPN).
  • FPN Feature Pyramid Networks
  • the executive body can use the convolutional neural network in the image recognition model to extract multi-layer features, and further use FPN to effectively fuse semantic information and positioning information to obtain fusion features, and input the fusion features to Multi-category detector to obtain recognition results of multi-card card images.
  • FPN is a method that uses a conventional convolutional neural network (Convolutional Neural Network, CNN) model to efficiently extract features of each dimension in a picture.
  • CNN convolutional Neural Network
  • FPN proposes a method that can effectively generate multi-dimensional feature representations of the same scale image in a single image view by using the feature expression structure of different dimensions of the same scale image from bottom to top in the conventional CNN model. It can effectively empower the conventional CNN model, so that feature maps with more expressive capabilities can be generated for the next stage of computer vision tasks.
  • the image recognition model is obtained based on the convolutional neural network and the feature pyramid network FPN, which can effectively improve the accuracy of the first recognition result corresponding to each of the at least two card images obtained, thereby improving the accuracy of the second recognition result. 2. Accuracy of recognition results.
  • Step 203 according to the category indicated by the first recognition result, perform a corresponding recognition operation on the card image corresponding to the first recognition result to obtain a second recognition result.
  • the execution subject after obtaining the first recognition result of each card image, the execution subject can input each card image into a different recognition branch according to the type of the card image indicated by the first recognition result, so as to execute According to the corresponding recognition operation, the second recognition result of the image of each card is obtained.
  • the recognition operation is used to indicate the operation of recognizing the field position and content in the card image.
  • the executive body can adopt field recognition technology in related technologies or future development technologies, for example, LSTM (Long Short Term Memory, long-term short-term memory network) + CTC (Connectionist temporal classification, temporal classification based on neural network), CRNN (Convolutional Recurrent Neural Network, convolutional cyclic neural network), etc., to identify the fields in the card image
  • LSTM Long Short Term Memory, long-term short-term memory network
  • CTC Connectionist temporal classification, temporal classification based on neural network
  • CRNN Convolutional Recurrent Neural Network, convolutional cyclic neural network
  • a corresponding recognition operation is performed on the card image corresponding to the first recognition result to obtain a second recognition result, including: responding to determining the first recognition result is the card home page, respectively, based on the field recognition model and the attention recognition model, the card image corresponding to the first recognition result is recognized to obtain the first sub-recognition result and the second sub-recognition result; the first sub-recognition result and the second sub-recognition result The two recognition results are fused to obtain a second recognition result.
  • the execution subject after the execution subject determines that the first recognition result is the card home page, it can recognize the card home page image according to the field recognition model and the attention recognition model respectively, and obtain the first sub-recognition result and the second sub-recognition result .
  • the attention recognition model is used to perform field recognition on images for which the position of the field area has not been determined.
  • the attention recognition model may be an RNN model that incorporates an attention model attention model.
  • the model has a higher accuracy rate for number recognition, that is, a higher accuracy rate for date and number recognition.
  • the execution subject inputs the card image corresponding to the first recognition result, for example, the driver's license homepage image or the driver's license homepage image, into the simultaneous field recognition model and the attention recognition model to obtain the first sub-recognition result and the second sub-recognition result.
  • the first recognition result for example, the driver's license homepage image or the driver's license homepage image
  • the execution subject After the execution subject obtains the first sub-recognition result and the second sub-recognition result, since the accuracy of the second sub-recognition result for digital recognition is relatively high, the execution subject can check the digits in the first sub-recognition result according to the second sub-recognition result. The numbers are corrected to obtain the second recognition result.
  • the first sub-recognition result and the second sub-recognition result are obtained.
  • Recognition result the first sub-recognition result and the second sub-recognition result are fused to obtain the second recognition result of the card image, which effectively improves the accuracy of the obtained multi-card image recognition result.
  • a corresponding recognition operation is performed on the card image corresponding to the first recognition result to obtain the second recognition result, including: responding to determining the first recognition result
  • the result is a card subpage, and the card image corresponding to the first recognition result is recognized based on the attention recognition model to obtain a second recognition result.
  • the execution subject determines that the first recognition result is the supplementary page of the card, it can further combine the type information of the card image. For pages containing more digital information, the card image corresponding to the first recognition result can be recognized based on the attention recognition model to obtain a second recognition result.
  • the attention recognition model may be an RNN model that incorporates an attention model attention model.
  • This model has a higher accuracy rate for identifying numbers, that is, it has a higher accuracy rate for identifying information such as dates and numbers.
  • the execution subject can input the card image, for example, the driver's license side page image, into the attention recognition model, obtain the recognition results of each field in the whole image, and determine the recognition result as the second recognition result.
  • the card image for example, the driver's license side page image
  • the card image corresponding to the first recognition result is recognized to obtain the second recognition result, which effectively improves the obtained multiple While improving the accuracy of card image recognition results, the recognition efficiency is improved.
  • a corresponding recognition operation is performed on the card image corresponding to the first recognition result to obtain the second recognition result, including: responding to determining the first recognition result
  • the result is a card subpage, and the card image corresponding to the first recognition result is recognized based on the field recognition model to obtain a second recognition result.
  • the execution subject determines that the first recognition result is the subpage of the card card, it can further combine the type information of the card image. If the subpage contains more text information, the card image corresponding to the first recognition result can be recognized based on the field recognition model to obtain the second recognition result.
  • the execution subject can input the card image, for example, the side page image of the driving license, into the field recognition model, obtain the recognition results of each field, and determine the recognition result as the second recognition result.
  • the second recognition result of the card image is obtained by recognizing the card image based on the field recognition model in response to determining that the first recognition result is a card subpage, which effectively improves the accuracy of the recognition result. At the same time, the recognition efficiency is improved.
  • the field recognition model includes an ROI perspective transformation processing unit.
  • the ROI (Region of Interest, Region of Interest) perspective transformation processing unit is used to perform perspective transformation processing on the text area in the image.
  • the perspective transformation processing is equivalent to Rotate and segment the text area to obtain multiple areas with fixed height and variable length, so that irregular text such as backlogs and overlaps in the image can be recognized.
  • This implementation helps to improve the accuracy of the obtained second recognition result by setting the ROI perspective transformation processing unit in the field recognition model.
  • the card image corresponding to the first recognition result is recognized based on the field recognition model to obtain the second recognition result, including: inputting the card image corresponding to the first recognition result into the field recognition
  • the area detection unit detects and outputs the position of the text area in the card image
  • the ROI perspective transformation processing unit obtains the features of the text area according to the position of the text area, and performs perspective transformation processing on the features of the text area to obtain
  • the text recognition unit recognizes the text content included in the text area to obtain a second recognition result according to the aligned ROI features based on the spatial attention mechanism.
  • the field recognition model may further include: an area detection unit and a character recognition unit in addition to the interest region perspective transformation processing unit.
  • the area detection unit is used to output the position of the character area in the image.
  • the most commonly used representation of the text area can be represented by a quadrilateral.
  • the area detection unit can be based on the full convolution operation, using the position of the four corner points to directly predict the deviation coordinates, the predicted position obtained through the transformation process forms a quadrilateral text area, and the final candidate quadrilateral frame is obtained through the non-maximum value suppression algorithm.
  • the execution subject can calculate the field line candidate frame according to the extracted global features of the card image, realize the prediction of the text position of the field line and the corner point of the bounding box to determine the position of the text area.
  • the executive body can first input the card image into the full convolutional network, and the final output is a 9-channel feature map, one of which is the confidence level of whether each pixel position in the picture is a text, and the remaining 8 channels indicate whether the pixel position is If it is text, the position corresponds to the x and y coordinate offsets ( ⁇ x1, ⁇ y1, ⁇ x2, ⁇ y2, ⁇ x3, ⁇ y3, ⁇ x4, ⁇ y4) of the four corners of the text bounding box.
  • NMS non-maximum suppression
  • the ROI perspective transformation processing unit performs ROI transformation on the image with the position of the text area determined, that is, the image with the position of the text area is determined through affine transformation It is transformed into region-of-interest features of a uniform scale for subsequent text recognition processing.
  • the text recognition unit is used to generate a recognized character sequence result according to the ROI feature processed by the ROI perspective transformation processing unit, that is, to recognize the text content included in the text area to obtain the second recognition result.
  • the feature map in the bounding box is transformed to the feature map F' (region of interest feature) with fixed height and width, while maintaining Keeping the aspect ratio unchanged, use (W,H,C) to represent the dimension of F'.
  • the text recognition unit can be realized by using a text recognition model in the existing technology or in the future development technology, for example, a CTC (Connectionist temporal classification, temporal classification based on a neural network) model, a Seq2Seq model, etc., which are not limited in the present disclosure.
  • a CTC Connectionist temporal classification, temporal classification based on a neural network
  • Seq2Seq model etc., which are not limited in the present disclosure.
  • the execution subject can use a sequence-to-sequence model (seq2seq) for text recognition.
  • This module consists of an RNN encoder and an RNN decoder.
  • the feature map F' region of interest features
  • the feature of this step is the flattening of the feature of F' at this step, and the feature The dimension is (H*C).
  • the time series is passed through the RNN encoder to obtain encoded features.
  • the decoder is another RNN model, which receives the character encoding (char embedding) obtained in the previous decoding step and the context vector (context vector) obtained in the previous decoding step at each decoding time step, and outputs the character prediction distribution of the decoding step; This goes on and on until the output at a certain moment is the end symbol ( ⁇ s>), that is, the decoding is stopped.
  • the input at the 0th moment of decoding is the preset start code ( ⁇ s>), and the context vector is obtained by the attention mechanism.
  • the detailed algorithm is given the state h of the hidden layer of the decoder, and calculates h and the encoding feature at each moment
  • the similarity of all coding moments is normalized by softmax, and then the coding features are weighted and averaged by the normalized similarity features, and the averaged features are the context vectors. Perform text recognition based on the context vector to obtain a second recognition result.
  • the card image corresponding to the first recognition result is input into the field recognition model, and the position of the text area in the card image is detected by the area detection unit and output.
  • the position acquires the features of the text area, and performs perspective transformation processing on the features of the text area to obtain the aligned area of interest features.
  • the text recognition unit Based on the spatial attention mechanism, the text recognition unit recognizes the features included in the text area according to the aligned area of interest features.
  • the text content is used to obtain the second recognition result, which helps to further improve the accuracy of the obtained second recognition result.
  • Step 204 summarizing the second recognition results and outputting them.
  • the executive body after the executive body completes the recognition sub-process of each type of card image and obtains the second recognition result of each card image, it can maintain the same number of cards by category as the number of card images for storing each card.
  • the array of the second recognition result of the image is used to summarize the second recognition results of each card image in the image to be recognized.
  • the execution subject may directly output the aggregated second recognition results, or may adjust the sequence of the aggregated second recognition results according to the position information of at least two card images in the image to be recognized and then output.
  • FIG. 3 is a schematic diagram of an application scenario of the image recognition method according to this embodiment.
  • Executing subject 301 obtains image to be recognized 302, includes at least two card images in the image to be recognized, for example, driver's license home page image, driver's license sub-page image, driving license home page image and driving license sub-page image; the image to be recognized is input
  • a preset image recognition model 303 to obtain a first recognition result 304, 305, 306, 307 corresponding to each card image in at least two card images, wherein the first recognition result is used to indicate the category of the card image , for example, driver's license homepage, driver's license supplementary page, driving license homepage, driving license supplementary page; according to the category indicated by the first recognition result 304, 305, 306, 307, execute the card image corresponding to the first recognition result
  • Corresponding recognition operations 308 , 309 , 310 , 311 obtain second recognition results 312 , 313 , 314 , 315 ; second recognition results 316 are
  • the image to be recognized includes at least two card images; inputting the image to be recognized into the preset image recognition model, and obtaining the A first recognition result corresponding to each card image, the first recognition result is used to indicate the category of the card image; according to the category indicated by the first recognition result, corresponding recognition is performed on the card image corresponding to the first recognition result operation to obtain the second recognition result; the second recognition results are summarized and output to realize the recognition of multi-card images, and at the same time perform different recognition operations on different cards, effectively improving the accuracy of multi-card image recognition and efficiency.
  • FIG. 4 it shows a flow 400 of another embodiment of the image recognition method shown in FIG. 2 .
  • the process 400 of the image recognition method may include the following steps:
  • Step 401 acquire an image to be recognized.
  • step 401 for implementation details and technical effects of step 401, reference may be made to the description of step 201, and details are not repeated here.
  • Step 402 Input the image to be recognized into a preset image recognition model to obtain a first recognition result corresponding to each of the at least two card images.
  • step 402 for implementation details and technical effects of step 402, reference may be made to the description of step 202, which will not be repeated here.
  • Step 403 Perform a corresponding recognition operation on the card image corresponding to the first recognition result according to the category indicated by the first recognition result to obtain a second recognition result.
  • step 403 for implementation details and technical effects of step 403, reference may be made to the description of step 203, which will not be repeated here.
  • Step 404 summarizing the second recognition results and outputting them based on the position information of at least two card images in the images to be recognized.
  • the executive body can summarize the second recognition results and obtain the results of at least two card images in the images to be recognized. position information, and then output a summarized second recognition result in an order determined based on the position information of at least two card images in the image to be recognized.
  • the position information of the at least two card images may be arranged in any position, for example, from top to bottom, from left to right, etc., which is not limited in the present disclosure.
  • the image to be recognized includes four card images, and the four card images are arranged in order from top to bottom in the image to be recognized.
  • the above-mentioned embodiment of the present disclosure highlights that by summarizing the second recognition results and outputting based on the position information of at least two card images in the image to be recognized, it realizes the recognition of multiple card images.
  • the adjustment of the output order of the recognition results enables the output results to correspond to the order of the card images, which improves the order and standardization of the output recognition results.
  • the present disclosure provides an embodiment of an image recognition device, which corresponds to the method embodiment shown in FIG. 1 , and the device can specifically Used in various electronic equipment.
  • the image recognition device 500 of this embodiment includes: an acquisition module 501 , an input module 502 , a recognition module 503 and an output module 504 .
  • the obtaining module 501 may be configured to obtain the image to be recognized.
  • the classification module 502 may be configured to input the image to be recognized into a preset image recognition model, and obtain a first recognition result corresponding to each card image in the at least two card images.
  • the recognition module 503 may be configured to perform a corresponding recognition operation on the card image corresponding to the first recognition result according to the category indicated by the first recognition result to obtain a second recognition result.
  • the output module 504 may be configured to collect and output the second recognition results.
  • the output module is further configured to: summarize the second recognition results and output them based on the position information of at least two card images in the image to be recognized.
  • the identification module is further configured to: in response to determining that the first identification result is the card home page, identify the card corresponding to the first identification result based on the field identification model and the attention identification model respectively.
  • the identification image is recognized to obtain a first sub-recognition result and a second sub-recognition result.
  • the recognition module is further configured to: in response to determining that the first recognition result is a card sub-page, recognize the card image corresponding to the first recognition result based on the attention recognition model , to get the second recognition result.
  • the identification module is further configured to: in response to determining that the first identification result is a card secondary page, identify the card image corresponding to the first identification result based on a field identification model , to get the second recognition result.
  • the field recognition model includes an ROI perspective transformation processing unit, and the ROI perspective transformation processing unit is configured to perform perspective transformation processing on the text region in the image.
  • the field recognition model further includes an area detection unit and a character recognition unit
  • the recognition module is further configured to: input the card image corresponding to the first recognition result into the field recognition model , the position of the text area in the card image is detected by the area detection unit and output, and the perspective transformation processing unit of the region of interest obtains the characteristics of the text area according to the position of the text area, and performs perspective transformation processing on the features of the text area to obtain Based on the aligned ROI features, the text recognition unit recognizes the text content included in the text area to obtain a second recognition result according to the aligned ROI features based on the spatial attention mechanism.
  • the image recognition model is obtained based on a convolutional neural network and a feature pyramid network FPN.
  • the acquisition, storage and application of the user's personal information involved are in compliance with relevant laws and regulations, and do not violate public order and good customs.
  • the present disclosure also provides an electronic device, a readable storage medium, and a computer program product.
  • FIG. 6 it is a block diagram of an electronic device according to an image recognition method according to an embodiment of the present disclosure.
  • Electronic device 600 is a block diagram of an electronic device according to an image recognition method of an embodiment of the present disclosure.
  • Electronic device is intended to represent various forms of digital computers, such as laptops, desktops, workstations, personal digital assistants, servers, blade servers, mainframes, and other suitable computers.
  • Electronic devices may also represent various forms of mobile devices, such as personal digital processing, cellular telephones, smart phones, wearable devices, and other similar computing devices.
  • the components shown herein, their connections and relationships, and their functions, are by way of example only, and are not intended to limit implementations of the disclosure described and/or claimed herein.
  • the electronic device includes: one or more processors 601, memory 602, and interfaces for connecting various components, including high-speed interfaces and low-speed interfaces.
  • the various components are interconnected using different buses and can be mounted on a common motherboard or otherwise as desired.
  • the processor may process instructions executed within the electronic device, including instructions stored in or on the memory, to display graphical information of a GUI on an external input/output device such as a display device coupled to an interface.
  • multiple processors and/or multiple buses may be used with multiple memories and multiple memories, if desired.
  • multiple electronic devices may be connected, with each device providing some of the necessary operations (eg, as a server array, a set of blade servers, or a multi-processor system).
  • a processor 601 is taken as an example.
  • the memory 602 is a non-transitory computer-readable storage medium provided in the present disclosure.
  • the memory stores instructions executable by at least one processor, so that the at least one processor executes the image recognition method provided in the present disclosure.
  • the non-transitory computer-readable storage medium of the present disclosure stores computer instructions, and the computer instructions are used to cause a computer to execute the image recognition method provided by the present disclosure.
  • the memory 602 can be used to store non-transitory software programs, non-transitory computer-executable programs and modules, such as program instructions/modules corresponding to the image recognition method in the embodiments of the present disclosure (for example, attached The acquisition module 501, the input module 502, the identification module 503 and the output module 504 shown in FIG. 5).
  • the processor 601 executes various functional applications and data processing of the server by running non-transitory software programs, instructions and modules stored in the memory 602, that is, implements the image recognition method in the above method embodiments.
  • the memory 602 may include a program storage area and a data storage area, wherein the program storage area may store an operating system and an application program required by at least one function; the data storage area may store data created by use of an electronic device for face tracking, and the like.
  • the memory 602 may include a high-speed random access memory, and may also include a non-transitory memory, such as at least one magnetic disk storage device, a flash memory device, or other non-transitory solid-state storage devices.
  • the memory 602 may optionally include a memory set remotely relative to the processor 601, and these remote memories may be connected to the electronic device for lane marking detection through a network. Examples of the aforementioned networks include, but are not limited to, the Internet, intranets, local area networks, mobile communication networks, and combinations thereof.
  • the electronic device of the image recognition method may further include: an input device 603 and an output device 604 .
  • the processor 601, the memory 602, the input device 603, and the output device 604 may be connected through a bus or in other ways. In FIG. 6, connection through a bus is taken as an example.
  • the input device 603 can receive input digital or character information, and generate key signal input related to user settings and function control of the electronic equipment for lane line detection, such as touch screen, small keyboard, mouse, trackpad, touchpad, indicator stick, One or more input devices such as mouse buttons, trackballs, joysticks, etc.
  • the output device 604 may include a display device, an auxiliary lighting device (eg, LED), a tactile feedback device (eg, a vibration motor), and the like.
  • the display device may include, but is not limited to, a liquid crystal display (LCD), a light emitting diode (LED) display, and a plasma display. In some implementations, the display device may be a touch screen.
  • Various implementations of the systems and techniques described herein can be implemented in digital electronic circuitry, integrated circuit systems, application specific ASICs (application specific integrated circuits), computer hardware, firmware, software, and/or combinations thereof. These various embodiments may include being implemented in one or more computer programs executable and/or interpreted on a programmable system including at least one programmable processor, the programmable processor Can be special-purpose or general-purpose programmable processor, can receive data and instruction from storage system, at least one input device, and at least one output device, and transmit data and instruction to this storage system, this at least one input device, and this at least one output device an output device.
  • machine-readable medium and “computer-readable medium” refer to any computer program product, apparatus, and/or means for providing machine instructions and/or data to a programmable processor ( For example, magnetic disks, optical disks, memories, programmable logic devices (PLDs), including machine-readable media that receive machine instructions as machine-readable signals.
  • machine-readable signal refers to any signal used to provide machine instructions and/or data to a programmable processor.
  • the systems and techniques described herein can be implemented on a computer having a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user. ); and a keyboard and pointing device (eg, a mouse or a trackball) through which a user can provide input to the computer.
  • a display device e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor
  • a keyboard and pointing device eg, a mouse or a trackball
  • Other kinds of devices can also be used to provide interaction with the user; for example, the feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and can be in any form (including Acoustic input, speech input or, tactile input) to receive input from the user.
  • the systems and techniques described herein can be implemented in a computing system that includes back-end components (e.g., as a data server), or a computing system that includes middleware components (e.g., an application server), or a computing system that includes front-end components (e.g., as a a user computer having a graphical user interface or web browser through which a user can interact with embodiments of the systems and techniques described herein), or including such backend components, middleware components, Or any combination of front-end components in a computing system.
  • the components of the system can be interconnected by any form or medium of digital data communication, eg, a communication network. Examples of communication networks include: Local Area Network (LAN), Wide Area Network (WAN) and the Internet.
  • a computer system may include clients and servers.
  • Clients and servers are generally remote from each other and typically interact through a communication network.
  • the relationship of client and server arises by computer programs running on the respective computers and having a client-server relationship to each other.
  • the accuracy and efficiency of recognizing multiple card images are effectively improved.
  • steps may be reordered, added or deleted using the various forms of flow shown above.
  • each step described in the present application may be executed in parallel, sequentially, or in a different order, as long as the desired result of the technical solution disclosed in the present disclosure can be achieved, no limitation is imposed herein.

Landscapes

  • Engineering & Computer Science (AREA)
  • Data Mining & Analysis (AREA)
  • Theoretical Computer Science (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Artificial Intelligence (AREA)
  • Evolutionary Biology (AREA)
  • Evolutionary Computation (AREA)
  • Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Image Analysis (AREA)
  • Character Discrimination (AREA)

Abstract

本公开提供了图像识别方法和装置，涉及人工智能技术领域，具体涉及计算机视觉和深度学习技术领域，具体可用于光学字符识别OCR等场景。具体实现方案为：获取待识别图像；将待识别图像输入预设的图像识别模型，得到与至少两个卡证图像中每一卡证图像对应的第一识别结果；根据第一识别结果所指示的类别，对与第一识别结果对应的卡证图像执行相应的识别操作，得到第二识别结果；汇总第二识别结果并输出。该方式有效提高了对多卡证图像进行识别的准确率和效率。

Description

图像识别方法和装置
相关申请的交叉引用
本专利申请要求于2021年8月18日提交的、申请号为202110947890.7、发明名称为“图像识别方法和装置”的中国专利申请的优先权，该申请的全文以引用的方式并入本申请中。
技术领域
本公开涉及人工智能技术领域，具体涉及计算机视觉和深度学习技术领域，具体可用于光学字符识别(Optical Character Recognition，OCR)等场景，尤其涉及一种图像识别方法和装置。
背景技术
在公共事件的处理中，往往会涉及到多个卡证图像的识别。
发明内容
本公开实施例提供了一种图像识别方法、装置、设备以及存储介质。
第一方面，本公开实施例提供了一种图像识别方法，该方法包括：获取待识别图像，待识别图像中包括至少两个卡证图像；将待识别图像输入预设的图像识别模型，得到与至少两个卡证图像中每一卡证图像对应的第一识别结果，第一识别结果用于指示卡证图像的类别；根据第一识别结果所指示的类别，对与第一识别结果对应的卡证图像执行相应的识别操作，得到第二识别结果；汇总第二识别结果并输出。
第二方面，本公开实施例提供了一种图像识别装置，该装置包括：获取模块，被配置成获取待识别图像，待识别图像中包括至少两个卡证图像；输入模块，被配置成将待识别图像输入预设的图像识别模型，得到与至少两个卡证图像中每一卡证图像对应的第一识别结果，第一识别结果用于指示卡证图像的类别；识别模块，被配置成根据第一识别结果所指示的类别，对与第一识别结果对应的卡证图像执行相应的识别操作，得到第二识别结果；输出模块，被配置成汇总第二识别结果并输出。
第三方面，本公开实施例提供了一种电子设备，该电子设备包括一个或多个处理器；存储装置，其上存储有一个或多个程序，当一个或多个程序被该一个或多个处理器执行，使得一个或多个处理器实现如第一方面的任一实施例的图像识别方法。
第四方面，本公开实施例提供了一种计算机可读介质，其上存储有计算机程序，该程序被处理器执行时实现如第一方面的任一实施例的图像识别方法。
第五方面，本公开实施例提供了一种计算机程序产品，包括计算机程序，该计算机程序被处理器执行时实现如第一方面的任一实施例的图像识别方法。
应当理解，本部分所描述的内容并非旨在标识本公开的实施例的关键或重要特征，也不用于限制本公开的范围。本公开的其他特征将通过以下的说明书而变得容易理解。
附图说明
图1是本公开可以应用于其中的示例性系统架构图；
图2是根据本公开的图像识别方法的一个实施例的流程图；
图3是根据本公开的图像识别方法的一个应用场景的示意图；
图4是根据本公开的图像识别方法的又一个实施例的流程图；
图5是根据本公开的图像识别装置的一个实施例的示意图；
图6是适于用来实现本公开实施例的电子设备的计算机系统的结构示意图。
具体实施方式
以下结合附图对本公开的示范性实施例做出说明，其中包括本公开实施例的各种细节以助于理解，应当将它们认为仅仅是示范性的。因此，本领域普通技术人员应当认识到，可以对这里描述的实施例做出各种改变和修改，而不会背离本公开的范围和精神。同样，为了清楚和简明，以下的描述中省略了对公知功能和结构的描述。
需要说明的是，在不冲突的情况下，本公开中的实施例及实施例中的特征可以相互组合。下面将参考附图并结合实施例来详细说明本公开。
图1示出了可以应用本公开的图像识别方法的实施例的示例性系统架构100。
如图1所示，系统架构100可以包括终端设备101、102、103，网络104和 服务器105。网络104用以在终端设备101、102、103和服务器105之间提供通信链路的介质。网络104可以包括各种连接类型，例如有线、无线通信链路或者光纤电缆等等。
用户可以使用终端设备101、102、103通过网络104与服务器105交互，以接收或发送消息等。终端设备101、102、103上可以安装有各种通讯客户端应用，例如，图像识别类应用、通信类应用等。
终端设备101、102、103可以是硬件，也可以是软件。当终端设备101、102、103为硬件时，可以是具有显示屏的各种电子设备，包括但不限于手机和笔记本电脑。当终端设备101、102、103为软件时，可以安装在上述所列举的电子设备中。其可以实现成多个软件或软件模块(例如用来提供图像识别服务)，也可以实现成单个软件或软件模块。在此不做具体限定。
服务器105可以是提供各种服务的服务器，例如，获取待识别图像；将待识别图像输入预设的图像识别模型，得到与至少两个卡证图像中每一卡证图像对应的第一识别结果；根据第一识别结果所指示的类别，对与第一识别结果对应的卡证图像执行相应的识别操作，得到第二识别结果；汇总第二识别结果并输出。
需要说明的是，服务器105可以是硬件，也可以是软件。当服务器105为硬件时，可以实现成多个服务器组成的分布式服务器集群，也可以实现成单个服务器。当服务器为软件时，可以实现成多个软件或软件模块(例如用来提供图像识别服务)，也可以实现成单个软件或软件模块。在此不做具体限定。
需要指出的是，本公开的实施例所提供的图像识别方法可以由服务器105执行，也可以由终端设备101、102、103执行，还可以由服务器105和终端设备101、102、103彼此配合执行。相应地，图像识别装置包括的各个部分(例如各个单元、子单元、模块、子模块)可以全部设置于服务器105中，也可以全部设置于终端设备101、102、103中，还可以分别设置于服务器105和终端设备101、102、103中。
应该理解，图1中的终端设备、网络和服务器的数目仅仅是示意性的。根据实现需要，可以具有任意数目的终端设备、网络和服务器。
图2示出了图像识别方法的实施例的流程示意图200。该图像识别方法包括以下步骤：
步骤201，获取待识别图像。
在本实施例中，执行主体(例如，图1中的服务器105或终端设备101、102、103)可以采用有线或无线方式从本地，如存放待识别图像的图像采集设备，或存放待识别图像的远端设备获取待识别图像。
其中，无线连接方式可以包括但不限于3G/4G连接、WiFi连接、蓝牙连接、WiMAX连接、Zigbee连接、UWB(ultra wideband)连接、以及其他现在已知或将来开发的无线连接方式。
这里，待识别图像包括至少两个卡证图像。
其中，卡证图像可以是任意卡证的图像，例如，驾驶证、行驶证、学生证、护照、社保卡等等，本公开对此不作限定。
需要指出的是，待识别图像可以是包括多个卡证图像的图像集，也可以是混贴有至少两个卡证图像的一幅图像，本公开对此不作限定。
具体地，待识别图像为一幅图像，图像中包括四个卡证图像，例如，分别为驾驶证主页、驾驶证副页、行驶证主页和行驶证副页的图像。
步骤202，将待识别图像输入预设的图像识别模型，得到与至少两个卡证图像中每一卡证图像对应的第一识别结果。
在本实施例中，执行主体在获取到待识别图像后，将待识别图像输入预设的图像识别模型，得到与至少两个卡证图像中每一卡证图像对应的第一识别结果，第一识别结果用于指示卡证图像的类别。
需要说明的是，卡证图像的类别可以包括卡证图像的类型信息和属性信息。这里，卡证图像的类型信息用于指示卡证图像的类型，例如，驾驶证、行驶证、学生证等，卡证图像的属性信息用于指示卡证图像的页面类别，例如，卡证主页、卡证副页等。
其中，图像识别模型可基于标注有卡证图像的类别标签的样本图像训练得到。
这里，图像识别模型可以基于现有技术或未来发展技术中的人工神经网络训练得到，例如，卷积神经网络、循环神经网络等，本公开对此不作限定。
具体地，待识别图像包括四个卡证图像，例如，分别为驾驶证主页图像、驾驶证副页图像、行驶证主页图像和行驶证副页图像。将待识别图像输入预设的图像识别模型，得到各卡证图像的第一识别结果，即得到待识别图像中驾驶证主页图像、驾驶证副页图像、行驶证主页图像和行驶证副页图像中每一图像的第一识别结果。
在一些可选的方式中，图像识别模型基于卷积神经网络和特征金字塔网络(Feature Pyramid Networks，FPN)得到。
在本实现方式中，对于多卡证识别来说，由于拍摄视角、距离等因素影响，会出现卡证大小和仿射形变，利用不同感受野特征融合的方式可以获取更丰富抽象的特征。
具体地，为了获取更加丰富的特征，执行主体可利用图像识别模型中的卷积神经网络提取多层特征，并进一步利用FPN将语义信息和定位信息进行有效融合，得到融合特征，将融合特征输入多分类检测器，得到多卡证图像的识别结果。
其中，FPN是一种利用常规卷积神经网络(Convolutional Neural Network，CNN)模型来高效提取图片中各维度特征的方法。FPN通过利用常规CNN模型内部从底至上各个层对同一scale图片不同维度的特征表达结构，提出了一种可有效在单一图片视图下生成对其的多维度特征表达的方法。它可以有效地赋能常规CNN模型，从而可以生成出表达能力更强的feature maps以供下一阶段计算机视觉任务使用。
该实现方式中图像识别模型基于卷积神经网络和特征金字塔网络FPN得到，可有效提升获取到的至少两个卡证图像中每一卡证图像对应的第一识别结果的准确性，进而提高第二识别结果的准确性。
步骤203，根据第一识别结果所指示的类别，对与第一识别结果对应的卡证图像执行相应的识别操作，得到第二识别结果。
在本实施例中，执行主体在获取到各卡证图像的第一识别结果后，可根据第一识别结果所指示的卡证图像的类别，将各卡证图像输入不同的识别分支，以执行相应的识别操作，得到各卡证的图像的第二识别结果。
其中，识别操作用于指示对卡证图像中的字段位置和内容进行识别的操作。
执行主体可以采用相关技术或未来发展技术中的字段识别技术，例如，LSTM(Long Short Term Memory，长短期记忆网络)+CTC(Connectionist temporal classification，基于神经网络的时序类分类)、CRNN(Convolutional Recurrent Neural Network，卷积循环神经网络)等，对卡证图像中的字段进行识别
在一些可选的方式中，根据第一识别结果所指示的类别，对与第一识别结果对应的卡证图像执行相应的识别操作，得到第二识别结果，包括：响应于确定第一识别结果为卡证主页，分别基于字段识别模型和attention识别模型对与该第一 识别结果对应的卡证图像进行识别，得到第一子识别结果和第二子识别结果；将第一子识别结果和第二子识别结果进行融合，得到第二识别结果。
在本实施例中，执行主体在确定第一识别结果为卡证主页后，可分别根据字段识别模型和attention识别模型对卡证主页图像进行识别，得到第一子识别结果和第二子识别结果。其中，attention识别模型用于对未确定出字段区域位置的图像进行字段识别。
这里，attention识别模型可以为融合了注意力模型attention model的RNN模型。该模型对数字进行识别的准确率较高，也即对日期、号码类识别的准确率较高。
具体地，执行主体将与第一识别结果对应的卡证图像，例如，驾驶证主页图像或行驶证主页图像输入同时字段识别模型和attention识别模型，以获取第一子识别结果和第二子识别结果。
执行主体在获取到第一子识别结果和第二子识别结果后，由于第二子识别结果对数字识别的准确率较高，执行主体可根据第二子识别结果对第一子识别结果中的数字进行校正，得到第二识别结果。
该实现方式通过响应于确定第一识别结果为卡证主页，分别基于字段识别模型和attention识别模型对与该第一识别结果对应的卡证图像进行识别，得到第一子识别结果和第二子识别结果；将第一子识别结果和第二子识别结果进行融合，得到该卡证图像的第二识别结果，有效提升了获取到的多卡证图像识别结果的准确率。
在一些可选的实现方式中，根据第一识别结果所指示的类别，对与第一识别结果对应的卡证图像执行相应的识别操作，得到第二识别结果，包括：响应于确定第一识别结果为卡证副页，基于attention识别模型对与该第一识别结果对应的卡证图像进行识别，得到第二识别结果。
在本实现方式中，执行主体在确定第一识别结果为卡证副页后，可进一步结合卡证图像的类型信息，若卡证的类型信息指示的类型为驾驶证、港澳通行证等，其副页包含较多数字信息的类型，则可基于attention识别模型对与该第一识别结果对应的卡证图像进行识别，得到第二识别结果。
其中，attention识别模型可以为融合了注意力模型attention model的RNN模型。该模型对数字进行识别的准确率较高，也即对日期、号码类信息识别的准确 率较高。
具体地，执行主体可将卡证图像，例如，驾驶证副页图像，输入attention识别模型，获取整图各字段识别结果，并将识别结果确定为第二识别结果。
该实现方式通过响应于确定第一识别结果为卡证副页，基于attention识别模型对与该第一识别结果对应的卡证图像进行识别，得到第二识别结果，在有效提升了获取到的多卡证图像识别结果的准确率的同时，提高了识别效率。
在一些可选的实现方式中，根据第一识别结果所指示的类别，对与第一识别结果对应的卡证图像执行相应的识别操作，得到第二识别结果，包括：响应于确定第一识别结果为卡证副页，基于字段识别模型对与该第一识别结果对应的卡证图像进行识别，得到第二识别结果。
在本实现方式中，执行主体在确定第一识别结果为卡证副页后，可进一步结和卡证图像的类型信息，若卡证的类型信息指示的类型为行驶证、学生证等，其副页包含较多文字信息的类型，则可基于字段识别模型对与该第一识别结果对应的卡证图像进行识别，得到第二识别结果。
具体地，执行主体可将卡证图像，例如，行驶证副页图像，输入字段识别模型，获取各字段识别结果，并将识别结果确定为第二识别结果。
该实现方式通过响应于确定第一识别结果为卡证副页，基于字段识别模型对该卡证图像进行识别，得到该卡证图像的第二识别结果，在有效提升了识别结果的准确率的同时，提高了识别效率。
在一些可选的方式中，字段识别模型包括感兴趣区域透视变换处理单元。
在本实现方式中，感兴趣区域透视变换处理单元，即ROI(Region of Interest，感兴趣区域)透视变换处理单元，该单元用于对图像中的文字区域进行透视变换处理，该透视变换处理相当于对文字区域进行旋转、分割等处理，得到多个高度固定，长度可变的区域，从而可以使得图像中存在的积压、重叠等不规则的文字能够被识别出来。
该实现方式通过在字段识别模型中设置感兴趣区域透视变换处理单元，有助于提升获取到的第二识别结果的准确性。
在一些可选的方式中，基于字段识别模型对与该第一识别结果对应的卡证图像进行识别，得到第二识别结果，包括：将与该第一识别结果对应的卡证图像输入字段识别模型中，由区域检测单元检测该卡证图像中文字区域的位置并输出， 感兴趣区域透视变换处理单元根据文字区域的位置获取文字区域的特征，并对文字区域的特征进行透视变换处理，得到对齐后的感兴趣区域特征，文字识别单元基于空间注意力机制，根据对齐后的所述感兴趣区域特征，识别文字区域包括的文字内容以获取第二识别结果。
在本实现方式中，字段识别模型除包括感兴趣区域透视变换处理单元外，还可以包括：区域检测单元和文字识别单元。
其中，区域检测单元用于输出图像中文字区域的位置。这里，文字区域最常用的表示方式可以采用四边形来表示。区域检测单元可以基于全卷积操作，采用四角点位置直接预测偏差坐标，经过变换处理得到的预测位置组成四边形的文字区域，经过非极大值抑制算法得到最终的候选四边形框的四个顶点的位置坐标。
具体地，执行主体可根据提取的卡证图像的全局特征计算字段行候选框，实现字段行文字位置和包围盒角点的预测以确定文字区域的位置。执行主体可首先将卡证图像输入全卷积网络，最后的输出为9通道的特征图，其中一个通道为图片中每个像素位置是否为文字的置信度，其余8个通道表示若该像素位置为文字，则该位置对应文字包围盒四角点的x，y坐标偏移量(Δx1,Δy1,Δx2,Δy2,Δx3,Δy3,Δx4,Δy4)。通过设置置信度阈值可提取出高置信度文字像素点位置(X,Y)，而后通过偏移图回归出文字候选的包围盒坐标(x1,y1,x2,y2,x3,y3,x4,y4)＝(X+Δx1,Y+Δy1,X+Δx2,Y+Δy2,X+Δx3,Y+Δy3,X+Δx4,Y+Δy4)。给定一张图上的文字候选，通过非极大值抑制(NMS)后可滤除重复检出的文字框，给出重复度较高的文字候选区域。将重复度较高的文字候选区域确定为图像中的文字区域位置。
进一步地，执行主体在确定出图像中的文字区域的位置后，由ROI透视变换处理单元对确定出文字区域的位置的图像进行ROI变换，也即将确定出文字区域的位置的图像通过仿射变换变换为统一尺度的感兴趣区域特征，以进行后续的文字识别处理。
这里，文字识别单元用于根据ROI透视变换处理单元处理的感兴趣区域特征，生成识别出的字符序列结果，即识别文字区域包括的文字内容以获取第二识别结果。
具体地，给定一个特征图F及一个包围盒四角点坐标，通过仿射变换，将包围盒内的特征图变换至定高变宽的特征图F’(感兴趣区域特征)上，同时保持长宽比不变，用(W,H,C)来表示F’的维度。
这里，文字识别单元可以采用现有技术或未来发展技术中的文本识别模型实现，例如，CTC(Connectionist temporal classification，基于神经网络的时序类分类)模型、Seq2Seq模型等，本公开对此不作限定。
具体地，执行主体可采用序列至序列模型(seq2seq)进行文字识别。该模块由一个RNN编码器及一个RNN解码器组成。首先将特征图F’(感兴趣区域特征)按列切片，组成一个时间序列，其中沿宽的每一列为一个编码时间步，该步的特征为F’在该步上特征的扁平化，特征维度为(H*C)。该时间序列通过RNN编码器，得到编码特征。解码器为另一个RNN模型，在每一个解码时间步接收上一解码步得到的字符编码(char embedding)以及上一解码步得到的上下文向量(context vector)，输出该解码步的字符预测分布；如此循环往复，直到某时刻的输出结果为结束符号(<\s>)，即停止解码。解码的第0个时刻输入为预设的开始符编码(<s>)，上下文向量由注意力机制得到，其详细算法为给定解码器隐含层状态h，计算h与编码特征每一时刻的相似度，通过softmax归一化所有编码时刻的相似度，而后将编码特征通过归一化的相似度特征进行加权平均，平均后的特征即为上下文向量。基于上下文向量进行文本识别，得到第二识别结果。
该实现方式通过将与该第一识别结果对应的卡证图像输入字段识别模型中，由区域检测单元检测该卡证图像中文字区域的位置并输出，感兴趣区域透视变换处理单元根据文字区域的位置获取文字区域的特征，并对文字区域的特征进行透视变换处理，得到对齐后的感兴趣区域特征，文字识别单元基于空间注意力机制，根据对齐后的感兴趣区域特征，识别文字区域包括的文字内容以获取第二识别结果，有助于进一步提升获取到的第二识别结果的准确性。
步骤204，汇总第二识别结果并输出。
在本实施例中，执行主体在完成各类别卡证图像的识别子流程后得到各卡证图像的第二识别结果之后，可按类别维护与卡证图像数量相同数量个用于存储各卡证图像第二识别结果的数组，以对待识别图像中各卡证图像的第二识别结果进行汇总。
进一步地，执行主体可将汇总的第二识别结果直接输出，也可根据待识别图像中至少两个卡证图像的位置信息对汇总的各第二识别结果的顺序进行调整后输出。
继续参见图3，图3是根据本实施例的图像识别方法的应用场景的一个示意 图。执行主体301获取待识别图像302，待识别图像中包括至少两个卡证图像，例如，驾驶证主页图像、驾驶证副页图像、行驶证主页图像和行驶证副页图像；将待识别图像输入预设的图像识别模型303，得到与至少两个卡证图像中每一卡证图像对应的第一识别结果304、305、306、307，其中，第一识别结果用于指示卡证图像的类别，例如，驾驶证主页、驾驶证副页、行驶证主页、行驶证副页；根据第一识别结果304、305、306、307所指示的类别，对与第一识别结果对应的卡证图像执行相应的识别操作308、309、310、311，得到第二识别结果312、313、314、315；汇总第二识别结果316并输出。
本公开的实施例提供的图像识别方法，通过获取待识别图像，待识别图像中包括至少两个卡证图像；将待识别图像输入预设的图像识别模型，得到与至少两个卡证图像中每一卡证图像对应的第一识别结果，第一识别结果用于指示卡证图像的类别；根据第一识别结果所指示的类别，对与第一识别结果对应的卡证图像执行相应的识别操作，得到第二识别结果；汇总第二识别结果并输出，实现了对多卡证图像的识别，同时对不同卡证执行不同的识别操作，有效提高了对多卡证图像进行识别的准确率和效率。
进一步参考图4，其示出了图2所示的图像识别方法的又一个实施例的流程400。在本实施例中，图像识别方法的流程400，可包括以下步骤：
步骤401，获取待识别图像。
在本实施例中，步骤401的实现细节和技术效果，可以参考对步骤201的描述，在此不再赘述。
步骤402，将待识别图像输入预设的图像识别模型，得到与至少两个卡证图像中每一卡证图像对应的第一识别结果。
在本实施例中，步骤402的实现细节和技术效果，可以参考对步骤202的描述，在此不再赘述。
步骤403，根据第一识别结果所指示的类别，对与该第一识别结果对应的卡证图像执行相应的识别操作，得到第二识别结果。
在本实施例中，步骤403的实现细节和技术效果，可以参考对步骤203的描述，在此不再赘述。
步骤404，汇总第二识别结果并基于待识别图像中至少两个卡证图像的位置信息进行输出。
在本实施例中，执行主体在获取到至少两个卡证图像中各卡证图像的第二识别结果后，可将各第二识别结果汇总并获取待识别图像中至少两个卡证图像的位置信息，进而按照基于待识别图像中至少两个卡证图像的位置信息确定的顺序输出汇总的第二识别结果。
这里，至少两个卡证图像的位置信息可以是任意的位置排列，例如，由上到下、由左到右等，本公开对此不作限定。
具体地，待识别图像中包括四个卡证图像，四个卡证图像在待识别图像中按照从上到下的顺序排列，例如，分别为驾驶证主页图像、驾驶证副页图像、行驶证主页图像、行驶证副页图像，并且执行主体在汇总各卡证图像的第二识别结果后，将驾驶证主页图像、驾驶证副页图像、行驶证主页图像和行驶证副页图像的第二识别结果分别按照从上到下的顺序进行输出。
本公开的上述实施例，与图2所示实施例相比，突出了通过汇总第二识别结果并基于待识别图像中至少两个卡证图像的位置信息进行输出，实现了对多卡证图像的识别结果的输出顺序的调整，使得输出结果可以与卡证图像的顺序相对应，提升了输出的识别结果的有序性和规范性。
进一步参考图5，作为对上述各图所示方法的实现，本公开提供了一种图像识别装置的一个实施例，该装置实施例与图1所示的方法实施例相对应，该装置具体可以应用于各种电子设备中。
如图5所示，本实施例的图像识别装置500包括：获取模块501、输入模块502、识别模块503和输出模块504。
其中，获取模块501，可被配置成获取待识别图像。
分类模块502，可被配置成将待识别图像输入预设的图像识别模型，得到与至少两个卡证图像中每一卡证图像对应的第一识别结果。
识别模块503，可被配置成根据第一识别结果所指示的类别，对与第一识别结果对应的卡证图像执行相应的识别操作，得到第二识别结果。
输出模块504，可被配置成汇总第二识别结果并输出。
在本实施例的一些可选的方式中，输出模块进一步被配置成：汇总第二识别结果并基于待识别图像中至少两个卡证图像的位置信息进行输出。
在本实施例的一些可选的方式中，识别模块进一步被配置成：响应于确定第一识别结果为卡证主页，分别基于字段识别模型和attention识别模型对与该第一 识别结果对应的卡证图像进行识别，得到第一子识别结果和第二子识别结果。
在本实施例的一些可选的方式中，识别模块进一步被配置成：响应于确定第一识别结果为卡证副页，基于attention识别模型对与该第一识别结果对应的卡证图像进行识别，得到第二识别结果。
在本实施例的一些可选的方式中，识别模块进一步被配置成：响应于确定第一识别结果为卡证副页，基于字段识别模型对与该第一识别结果对应的卡证图像进行识别，得到第二识别结果。
在本实施例的一些可选的方式中，字段识别模型包括感兴趣区域透视变换处理单元，兴趣区域透视变换处理单元用于对图像中的文字区域进行透视变换处理。
在本实施例的一些可选的方式中，字段识别模型还包括区域检测单元和文字识别单元，以及识别模块进一步被配置成：将与该第一识别结果对应的卡证图像输入字段识别模型中，由区域检测单元检测该卡证图像中文字区域的位置并输出，感兴趣区域透视变换处理单元根据文字区域的位置获取所述文字区域的特征，并对文字区域的特征进行透视变换处理，得到对齐后的感兴趣区域特征，文字识别单元基于空间注意力机制，根据对齐后的所述感兴趣区域特征，识别文字区域包括的文字内容以获取第二识别结果。
在本实施例的一些可选的方式中，图像识别模型基于卷积神经网络和特征金字塔网络FPN得到。
本公开的技术方案中，所涉及的用户个人信息的获取，存储和应用等，均符合相关法律法规的规定，且不违背公序良俗。
根据本公开的实施例，本公开还提供了一种电子设备、一种可读存储介质和一种计算机程序产品。
如图6所示，是根据本公开实施例的图像识别方法的电子设备的框图。
600是根据本公开实施例的图像识别方法的电子设备的框图。电子设备旨在表示各种形式的数字计算机，诸如，膝上型计算机、台式计算机、工作台、个人数字助理、服务器、刀片式服务器、大型计算机、和其它适合的计算机。电子设备还可以表示各种形式的移动装置，诸如，个人数字处理、蜂窝电话、智能电话、可穿戴设备和其它类似的计算装置。本文所示的部件、它们的连接和关系、以及它们的功能仅仅作为示例，并且不意在限制本文中描述的和/或者要求的本公开的实现。
如图6所示，该电子设备包括：一个或多个处理器601、存储器602，以及 用于连接各部件的接口，包括高速接口和低速接口。各个部件利用不同的总线互相连接，并且可以被安装在公共主板上或者根据需要以其它方式安装。处理器可以对在电子设备内执行的指令进行处理，包括存储在存储器中或者存储器上以在外部输入/输出装置(诸如，耦合至接口的显示设备)上显示GUI的图形信息的指令。在其它实施方式中，若需要，可以将多个处理器和/或多条总线与多个存储器和多个存储器一起使用。同样，可以连接多个电子设备，各个设备提供部分必要的操作(例如，作为服务器阵列、一组刀片式服务器、或者多处理器系统)。图6中以一个处理器601为例。
存储器602即为本公开所提供的非瞬时计算机可读存储介质。其中，所述存储器存储有可由至少一个处理器执行的指令，以使所述至少一个处理器执行本公开所提供的图像识别方法。本公开的非瞬时计算机可读存储介质存储计算机指令，该计算机指令用于使计算机执行本公开所提供的图像识别方法。
存储器602作为一种非瞬时计算机可读存储介质，可用于存储非瞬时软件程序、非瞬时计算机可执行程序以及模块，如本公开实施例中的图像识别方法对应的程序指令/模块(例如，附图5所示的获取模块501、输入模块502、识别模块503和输出模块504)。处理器601通过运行存储在存储器602中的非瞬时软件程序、指令以及模块，从而执行服务器的各种功能应用以及数据处理，即实现上述方法实施例中的图像识别方法。
存储器602可以包括存储程序区和存储数据区，其中，存储程序区可存储操作系统、至少一个功能所需要的应用程序；存储数据区可存储人脸跟踪的电子设备的使用所创建的数据等。此外，存储器602可以包括高速随机存取存储器，还可以包括非瞬时存储器，例如至少一个磁盘存储器件、闪存器件、或其他非瞬时固态存储器件。在一些实施例中，存储器602可选包括相对于处理器601远程设置的存储器，这些远程存储器可以通过网络连接至车道线检测的电子设备。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。
图像识别方法的电子设备还可以包括：输入装置603和输出装置604。处理器601、存储器602、输入装置603和输出装置604可以通过总线或者其他方式连接，图6中以通过总线连接为例。
输入装置603可接收输入的数字或字符信息，以及产生与车道线检测的电子设备的用户设置以及功能控制有关的键信号输入，例如触摸屏、小键盘、鼠标、 轨迹板、触摸板、指示杆、一个或者多个鼠标按钮、轨迹球、操纵杆等输入装置。输出装置604可以包括显示设备、辅助照明装置(例如，LED)和触觉反馈装置(例如，振动电机)等。该显示设备可以包括但不限于，液晶显示器(LCD)、发光二极管(LED)显示器和等离子体显示器。在一些实施方式中，显示设备可以是触摸屏。
此处描述的系统和技术的各种实施方式可以在数字电子电路系统、集成电路系统、专用ASIC(专用集成电路)、计算机硬件、固件、软件、和/或它们的组合中实现。这些各种实施方式可以包括：实施在一个或者多个计算机程序中，该一个或者多个计算机程序可在包括至少一个可编程处理器的可编程系统上执行和/或解释，该可编程处理器可以是专用或者通用可编程处理器，可以从存储系统、至少一个输入装置、和至少一个输出装置接收数据和指令，并且将数据和指令传输至该存储系统、该至少一个输入装置、和该至少一个输出装置。
这些计算程序(也称作程序、软件、软件应用、或者代码)包括可编程处理器的机器指令，并且可以利用高级过程和/或面向对象的编程语言、和/或汇编/机器语言来实施这些计算程序。如本文使用的，术语“机器可读介质”和“计算机可读介质”指的是用于将机器指令和/或数据提供给可编程处理器的任何计算机程序产品、设备、和/或装置(例如，磁盘、光盘、存储器、可编程逻辑装置(PLD))，包括，接收作为机器可读信号的机器指令的机器可读介质。术语“机器可读信号”指的是用于将机器指令和/或数据提供给可编程处理器的任何信号。
为了提供与用户的交互，可以在计算机上实施此处描述的系统和技术，该计算机具有：用于向用户显示信息的显示装置(例如，CRT(阴极射线管)或者LCD(液晶显示器)监视器)；以及键盘和指向装置(例如，鼠标或者轨迹球)，用户可以通过该键盘和该指向装置来将输入提供给计算机。其它种类的装置还可以用于提供与用户的交互；例如，提供给用户的反馈可以是任何形式的传感反馈(例如，视觉反馈、听觉反馈、或者触觉反馈)；并且可以用任何形式(包括声输入、语音输入或者、触觉输入)来接收来自用户的输入。
可以将此处描述的系统和技术实施在包括后台部件的计算系统(例如，作为数据服务器)、或者包括中间件部件的计算系统(例如，应用服务器)、或者包括前端部件的计算系统(例如，具有图形用户界面或者网络浏览器的用户计算机，用户可以通过该图形用户界面或者该网络浏览器来与此处描述的系统和技术的 实施方式交互)、或者包括这种后台部件、中间件部件、或者前端部件的任何组合的计算系统中。可以通过任何形式或者介质的数字数据通信(例如，通信网络)来将系统的部件相互连接。通信网络的示例包括：局域网(LAN)、广域网(WAN)和互联网。
计算机系统可以包括客户端和服务器。客户端和服务器一般远离彼此并且通常通过通信网络进行交互。通过在相应的计算机上运行并且彼此具有客户端-服务器关系的计算机程序来产生客户端和服务器的关系。
根据本公开实施例的技术方案，有效提高了对多卡证图像进行识别的准确率和效率。
应该理解，可以使用上面所示的各种形式的流程，重新排序、增加或删除步骤。例如，本发申请中记载的各步骤可以并行地执行也可以顺序地执行也可以不同的次序执行，只要能够实现本公开公开的技术方案所期望的结果，本文在此不进行限制。
上述具体实施方式，并不构成对本公开保护范围的限制。本领域技术人员应该明白的是，根据设计要求和其他因素，可以进行各种修改、组合、子组合和替代。任何在本公开的精神和原则之内所作的修改、等同替换和改进等，均应包含在本公开保护范围之内。

Claims (19)

  1. 一种图像识别方法，包括：
    获取待识别图像，所述待识别图像中包括至少两个卡证图像；
    将待识别图像输入预设的图像识别模型，得到与所述至少两个卡证图像中每一卡证图像对应的第一识别结果，所述第一识别结果用于指示卡证图像的类别；
    根据第一识别结果所指示的类别，对与第一识别结果对应的卡证图像执行相应的识别操作，得到第二识别结果；
    汇总第二识别结果并输出。
  2. 根据权利要求1所述的方法，其中，所述汇总第二识别结果并输出，包括：
    汇总第二识别结果并基于所述待识别图像中至少两个卡证图像的位置信息进行输出。
  3. 根据权利要求1或2任一所述的方法，其中，所述根据第一识别结果所指示的类别，对与第一识别结果对应的卡证图像执行相应的识别操作，得到第二识别结果，包括：
    响应于确定第一识别结果为卡证主页，分别基于字段识别模型和attention识别模型对与该第一识别结果对应的卡证图像进行识别，得到第一子识别结果和第二子识别结果；
    将第一子识别结果和第二子识别结果进行融合，得到第二识别结果。
  4. 根据权利要求1或2任一所述的方法，其中，所述根据第一识别结果所指示的类别，对与第一识别结果对应的卡证图像执行相应的识别操作，得到第二识别结果，包括：
    响应于确定第一识别结果为卡证副页，基于attention识别模型对与该第一识别结果对应的卡证图像进行识别，得到第二识别结果。
  5. 根据权利要求1或2任一所述的方法，其中，所述根据第一识别结果所 指示的类别，对与第一识别结果对应的卡证图像执行相应的识别操作，得到第二识别结果，包括：
    响应于确定第一识别结果为卡证副页，基于字段识别模型对与该第一识别结果对应的卡证图像进行识别，得到第二识别结果。
  6. 根据权利要求5所述的方法，其中，所述字段识别模型包括感兴趣区域透视变换处理单元，所述兴趣区域透视变换处理单元用于对图像中的文字区域进行透视变换处理。
  7. 根据权利要求6所述的方法，其中，所述字段识别模型还包括区域检测单元和文字识别单元，以及基于字段识别模型对与该第一识别结果对应的卡证图像进行识别，得到第二识别结果，包括：
    将与该第一识别结果对应的卡证图像输入字段识别模型中，由所述区域检测单元检测该卡证图像中文字区域的位置并输出，所述感兴趣区域透视变换处理单元根据所述文字区域的位置获取所述文字区域的特征，并对所述文字区域的特征进行透视变换处理，得到对齐后的感兴趣区域特征，所述文字识别单元基于空间注意力机制，根据对齐后的所述感兴趣区域特征，识别所述文字区域包括的文字内容以获取第二识别结果。
  8. 根据权利要求1或2任一所述的方法，其中，所述图像识别模型基于卷积神经网络和特征金字塔网络FPN得到。
  9. 一种图像识别装置，包括：
    获取模块，被配置成获取待识别图像，所述待识别图像中包括至少两个卡证图像；
    输入模块，被配置成将待识别图像输入预设的图像识别模型，得到与所述至少两个卡证图像中每一卡证图像对应的第一识别结果，所述第一识别结果用于指示卡证图像的类别；
    识别模块，被配置成根据第一识别结果所指示的类别，对与第一识别结果对应的卡证图像执行相应的识别操作，得到第二识别结果；
    输出模块，被配置成汇总第二识别结果并输出。
  10. 根据权利要求9所述的装置，其中，所述输出模块进一步被配置成：
    汇总第二识别结果并基于所述待识别图像中至少两个卡证图像的位置信息进行输出。
  11. 根据权利要求9或10任一所述的装置，其中，所述识别模块进一步被配置成：
    响应于确定第一识别结果为卡证主页，分别基于字段识别模型和attention识别模型对与该第一识别结果对应的卡证图像进行识别，得到第一子识别结果和第二子识别结果；
    将第一子识别结果和第二子识别结果进行融合，得到第二识别结果。
  12. 根据权利要求9或10任一所述的装置，其中，所述识别模块进一步被配置成：
    响应于确定第一识别结果为卡证副页，基于attention识别模型对与该第一识别结果对应的卡证图像进行识别，得到第二识别结果。
  13. 根据权利要求9或10任一所述的装置，其中，所述识别模块进一步被配置成：
    响应于确定第一识别结果为卡证副页，基于字段识别模型对与该第一识别结果对应的卡证图像进行识别，得到第二识别结果。
  14. 根据权利要求13所述的装置，其中，所述字段识别模型包括感兴趣区域透视变换处理单元，所述兴趣区域透视变换处理单元用于对图像中的文字区域进行透视变换处理。
  15. 根据权利要求14所述的装置，其中，所述字段识别模型还包括区域检测单元和文字识别单元，以及所述识别模块进一步被配置成：
    将与该第一识别结果对应的卡证图像输入字段识别模型中，由所述区域检测 单元检测该卡证图像中文字区域的位置并输出，所述感兴趣区域透视变换处理单元根据所述文字区域的位置获取所述文字区域的特征，并对所述文字区域的特征进行透视变换处理，得到对齐后的感兴趣区域特征，所述文字识别单元基于空间注意力机制，根据对齐后的所述感兴趣区域特征，识别所述文字区域包括的文字内容以获取第二识别结果。
  16. 根据权利要求9或10任一所述的装置，其中，所述图像识别模型基于卷积神经网络和特征金字塔网络FPN得到。
  17. 一种电子设备，其特征在于，包括：
    至少一个处理器；以及
    与所述至少一个处理器通信连接的存储器；其中，
    所述存储器存储有可被所述至少一个处理器执行，以使所述至少一个处理器能够执行权利要求1-8中任一项所述的方法。
  18. 一种非瞬时计算机可读存储介质，存储有计算机指令，其中，所述计算机指令用于使所述计算机执行权利要求1-8中任一项所述的方法。
  19. 一种计算机程序产品，包括计算机程序，其中，所述计算机程序在被处理器执行时实现根据权利要求1-8中任一项所述的方法。
PCT/CN2022/106160 2021-08-18 2022-07-18 图像识别方法和装置 WO2023020176A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202110947890.7A CN113657398B (zh) 2021-08-18 2021-08-18 图像识别方法和装置
CN202110947890.7 2021-08-18

Publications (1)

Publication Number Publication Date
WO2023020176A1 true WO2023020176A1 (zh) 2023-02-23

Family

ID=78480894

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2022/106160 WO2023020176A1 (zh) 2021-08-18 2022-07-18 图像识别方法和装置

Country Status (2)

Country Link
CN (1) CN113657398B (zh)
WO (1) WO2023020176A1 (zh)

Families Citing this family (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113657398B (zh) * 2021-08-18 2023-02-07 北京百度网讯科技有限公司 图像识别方法和装置
CN117975466A (zh) * 2024-04-01 2024-05-03 山东浪潮科学研究院有限公司 一种基于版面分析的通用场景卡证识别系统

Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20150341370A1 (en) * 2014-02-25 2015-11-26 Sal Khan Systems and methods relating to the authenticity and verification of photographic identity documents
CN109492643A (zh) * 2018-10-11 2019-03-19 平安科技（深圳）有限公司 基于ocr的证件识别方法、装置、计算机设备及存储介质
CN109657673A (zh) * 2017-10-11 2019-04-19 阿里巴巴集团控股有限公司 图像识别方法和终端
CN111275102A (zh) * 2020-01-19 2020-06-12 深圳壹账通智能科技有限公司 多证件类型同步检测方法、装置、计算机设备及存储介质
CN112287923A (zh) * 2020-12-24 2021-01-29 德联易控科技(北京)有限公司 卡证信息识别方法、装置、设备及存储介质
CN113657398A (zh) * 2021-08-18 2021-11-16 北京百度网讯科技有限公司 图像识别方法和装置
CN114445843A (zh) * 2022-01-25 2022-05-06 京东科技控股股份有限公司 固定版式的卡证图像文字识别方法和装置

Family Cites Families (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109815976A (zh) * 2018-12-14 2019-05-28 深圳壹账通智能科技有限公司 一种证件信息识别方法、装置及设备
CN110458011A (zh) * 2019-07-05 2019-11-15 北京百度网讯科技有限公司 端到端的文字识别方法及装置、计算机设备及可读介质
CN111242124B (zh) * 2020-01-13 2023-10-31 支付宝实验室(新加坡)有限公司 一种证件分类方法、装置及设备

Patent Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20150341370A1 (en) * 2014-02-25 2015-11-26 Sal Khan Systems and methods relating to the authenticity and verification of photographic identity documents
CN109657673A (zh) * 2017-10-11 2019-04-19 阿里巴巴集团控股有限公司 图像识别方法和终端
CN109492643A (zh) * 2018-10-11 2019-03-19 平安科技（深圳）有限公司 基于ocr的证件识别方法、装置、计算机设备及存储介质
CN111275102A (zh) * 2020-01-19 2020-06-12 深圳壹账通智能科技有限公司 多证件类型同步检测方法、装置、计算机设备及存储介质
CN112287923A (zh) * 2020-12-24 2021-01-29 德联易控科技(北京)有限公司 卡证信息识别方法、装置、设备及存储介质
CN113657398A (zh) * 2021-08-18 2021-11-16 北京百度网讯科技有限公司 图像识别方法和装置
CN114445843A (zh) * 2022-01-25 2022-05-06 京东科技控股股份有限公司 固定版式的卡证图像文字识别方法和装置

Also Published As

Publication number Publication date
CN113657398A (zh) 2021-11-16
CN113657398B (zh) 2023-02-07

Similar Documents

Publication Publication Date Title
US11854246B2 (en) Method, apparatus, device and storage medium for recognizing bill image
WO2023015941A1 (zh) 文本检测模型的训练方法和检测文本方法、装置和设备
KR102597377B1 (ko) 이미지 인식방법, 장치, 기기, 컴퓨터 저장매체 및 컴퓨터 프로그램
WO2023020176A1 (zh) 图像识别方法和装置
US11430265B2 (en) Video-based human behavior recognition method, apparatus, device and storage medium
EP3812963B1 (en) Vehicle re-identification method, apparatus, device and storage medium
CN111950292B (zh) 文本纠错模型的训练方法、文本纠错处理方法和装置
US11756332B2 (en) Image recognition method, apparatus, device, and computer storage medium
US11854237B2 (en) Human body identification method, electronic device and storage medium
US20210295472A1 (en) Method and apparatus for recognizing abnormal license plate, device and readable storage medium
US11468655B2 (en) Method and apparatus for extracting information, device and storage medium
US11403799B2 (en) Method and apparatus for recognizing face-swap, device and computer readable storage medium
CN111814637A (zh) 一种危险驾驶行为识别方法、装置、电子设备及存储介质
CN114863437B (zh) 文本识别方法、装置、电子设备和存储介质
US11810333B2 (en) Method and apparatus for generating image of webpage content
US20230260306A1 (en) Method and Apparatus for Recognizing Document Image, Storage Medium and Electronic Device
CN115422389B (zh) 处理文本图像的方法及装置、神经网络的训练方法
CN115578735B (zh) 文本检测方法和文本检测模型的训练方法、装置
US20230116040A1 (en) Method for recognizing dangerous driving behavior, electronic device and storage medium
WO2021159843A1 (zh) 对象识别方法和装置、电子设备、存储介质
US20230027813A1 (en) Object detecting method, electronic device and storage medium
CN113887394A (zh) 一种图像处理方法、装置、设备及存储介质
CN113688803B (zh) 公式识别方法、装置、电子设备及存储介质
CN116311271B (zh) 文本图像的处理方法及装置
CN116168442B (zh) 样本图像生成方法、模型的训练方法、目标检测方法

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: 22857489

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE