Text detection and recognition method, device and equipment for automobile certificate of conformity

Abstract

The application discloses a text detection and recognition method, device and equipment for automobile factory qualified certificates, and belongs to the technical field of image processing. The method comprises the following steps: acquiring digital original images of a plurality of automobile factory qualified certificates as initial data sets; performing image preprocessing on each image in the initial data sets to remove grid lines in the images and fade seals in the images, so as to obtain a preprocessed image data set; dividing the preprocessed image data set into a training set and a test set, and labeling the training set; constructing an improved differentiable binarization network, and training the network by using the training set; and inputting the test set into the improved differentiable binarization network to obtain a text detection and recognition result. Compared with a traditional DBNet text detection and recognition method, the application has more accurate detection and recognition results and stronger robustness in the scene of text detection and recognition of automobile factory qualified certificates.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of image processing, and in particular to a text detection and recognition method, device and equipment for vehicle factory qualified certificates. BACKGROUND

[0002] The recognition and collation of vehicle factory qualified certificates aims to guarantee the accuracy and legality of vehicle data in the vehicle production stage. These data contain information such as vehicle identification number and engine number, which is of great significance to the registration and management of information when the vehicle is delivered. However, with the rapid growth of automobile production, the number of verifications also increases accordingly. The current manual collation method is not only inefficient but also prone to errors, which has a great impact on the production process of vehicles.

[0003] There are conventional solutions for automatic detection and recognition of text, but the vehicle factory qualified certificate has features such as grid lines, seals, etc. that affect the accuracy of detection and recognition, as well as dense text, different fonts, etc. SUMMARY

[0004] The present application aims to overcome the shortcomings of the prior art and provide a text detection and recognition method, device and equipment for vehicle factory qualified certificates to reduce the impact of grid lines and seals on text detection and improve the accuracy of dense text recognition.

[0005] To achieve the above-mentioned purpose, the technical solution of the present application is:

[0006] In a first aspect, the present application provides a text detection and recognition method for vehicle factory qualified certificates, comprising:

[0007] Obtaining digital original images of a plurality of vehicle factory qualified certificates as initial data sets;

[0008] Performing image preprocessing on each image in the initial data set to remove grid lines in the image and fade seals in the image to obtain a preprocessed image data set;

[0009] Dividing the preprocessed image data set into a training set and a test set, and labeling the training set;

[0010] Constructing an improved differentiable binarization network and training it using the training set;

[0011] Inputting the test set into the improved differentiable binarization network to obtain a text detection and recognition result.

[0012] Optionally, the removal of grid lines in the image comprises:

[0013] The digital original image of the automobile factory qualified certificate is converted into a gray image, and a global threshold operation is performed on the gray image to obtain a binary image;

[0014] The binary image is inverted in pixel value;

[0015] The inverted binary image is analyzed and extracted in a connected region to obtain a grid line image;

[0016] The grid line image is dilated and eroded to obtain a region mask;

[0017] The gray image is inverted again in pixel, and is mixed with the region mask at the pixel level.

[0018] Optionally, the analysis and extraction of the inverted binary image in a connected region comprises:

[0019] In the inverted binary image, adjacent pixel points in the horizontal or vertical direction are attributed to the same connected region if they have the same pixel value;

[0020] Each connected region is assigned a unique label;

[0021] All connected regions are traversed based on the label, and the number of pixels contained in each connected region is calculated to determine the connected region with the most pixels;

[0022] The connected region with the most pixels is retained, and the remaining connected regions are removed.

[0023] Optionally, the conversion of the digital image of the automobile factory qualified certificate into a gray image and the global threshold operation on the gray image are represented as:

[0024]

[0025] In the formula, T represents the global threshold, f(x, y) is the set threshold, and g(x, y) is the changed threshold.

[0026] Optionally, the stamp in the faded image comprises:

[0027] The digital original image of the automobile factory qualified certificate is globally binaryzied;

[0028] The digital original image of the automobile factory qualified certificate is fused with the globally binaryzied image based on a weight;

[0029] The fused image is dilated and eroded.

[0030] Optionally, the differential binaryzied network is constructed by the following method:

[0031] In the backbone network ResNet18-vd, the first three 2D convolution kernels are replaced by deformable convolution kernels;

[0032] In the feature fusion FPN part, two attention mechanisms, SENet and CBAM, are introduced.

[0033] Optionally, the deformable convolution kernel is represented as:

[0034]

[0035] In the formula, P0 represents any point on the input feature map, w represents the weight of the corresponding position of the convolution kernel, P n represents the offset of each point in the convolution kernel relative to the center point, and ΔP n represents a new offset.

[0036] In a second aspect, the present application provides a text detection and recognition device for vehicle factory qualified certificates, the device comprising:

[0037] A digital image acquisition module is configured to acquire digital original images of a plurality of vehicle factory qualified certificates as an initial data set.

[0038] An image preprocessing module is configured to perform image preprocessing on each image in the initial data set to remove grid lines in the image and fade the seal in the image, thereby obtaining a preprocessed image data set.

[0039] A data set labeling module is configured to divide the preprocessed image data set into a training set and a test set, and label the training set.

[0040] An improved differentiable binarization network training module is configured to construct an improved differentiable binarization network and train the network using the training set.

[0041] A detection and recognition module is configured to input the test set into the improved differentiable binarization network to obtain a text detection and recognition result.

[0042] Optionally, the image preprocessing module comprises a grid line removal submodule, and the grid line removal submodule comprises:

[0043] A binarization image unit is configured to convert the digital original image of the vehicle factory qualified certificate into a grayscale image, and perform a global threshold operation on the grayscale image to obtain a binarization image.

[0044] An inversion unit is configured to invert the pixel values of the binarization image.

[0045] A grid line image unit is configured to analyze and extract the connected regions of the inverted binarization image to obtain a grid line image.

[0046] a processing unit configured to perform dilation and corrosion processing on the grid line image to obtain a region mask;

[0047] a mixing unit configured to perform pixel inversion on the gray scale image again and mix the pixel level with the region mask.

[0048] Optionally, the image preprocessing module further comprises a seal fading sub-module, and the seal fading sub-module comprises:

[0049] a global binarization unit configured to perform global binarization on the digitized original image of the vehicle factory certificate;

[0050] a fusion unit configured to fuse the digitized original image of the vehicle factory certificate and the image after global binarization based on a weight;

[0051] a processing unit configured to perform dilation and corrosion processing on the fused image.

[0052] In a third aspect, the present application provides an electronic device comprising a processor, a communication interface, a memory and a communication bus, wherein the processor, the communication interface and the memory complete mutual communication through the communication bus;

[0053] the memory is configured to store a computer program;

[0054] the processor is configured to execute the program stored on the memory to realize the steps of the text detection and recognition method for the vehicle factory certificate according to any one of claims 1-7

[0055] Compared with the prior art, the present application has the following beneficial effects:

[0056] (1) The present application uses connected region analysis on the basis of the traditional image processing method, which can remove the grid lines in the image, thereby avoiding the interference of the grid lines on the text detection and recognition;

[0057] (2) The present application proposes a method for fading the seal in the certificate, which can reduce the influence of the seal on the text and restore the clarity of the text covered by the seal;

[0058] (3) The present application uses a deformable convolution kernel instead of the original 2D convolution kernel, and introduces SENet and CBAM two attention mechanisms in the FPN structure, which improves the accuracy and robustness of the convolution network for text detection and recognition, thereby reducing the probability of text missed detection and improving the accuracy of text detection and recognition. BRIEF DESCRIPTION OF DRAWINGS

[0059] Figure 1 a flowchart of the method for text detection and recognition of the vehicle factory certificate provided by the embodiments of the present application;

[0060] Figure 2 Flowchart for removing grid lines in the image;

[0061] Figure 3 Flowchart for analyzing and extracting connected regions of the inverted binary image;

[0062] Figure 4 Flowchart for fading the seal in the image;

[0063] Figure 5 The overall flowchart of an application scenario for the method for detecting and recognizing the text of the automobile factory certificate provided by the embodiment of the present application;

[0064] Figure 6 The data set used in the application scenario of the method of the present application;

[0065] Figure 7 The specific steps and effects of removing grid lines in the application scenario of the method of the present application;

[0066] Figure 8 The specific steps and effects of fading the seal in the application scenario of the method of the present application;

[0067] Figure 9 The detection and recognition result of the conventional method;

[0068] Figure 10 The detection and recognition result of the method of the present application

[0069] Figure 11 Comparison of the detection and recognition results of the method of the present application and the conventional method; wherein a) is the detection and recognition result of the original DBNet, and b) is the detection and recognition result of the method of the present application;

[0070] Figure 12 The composition diagram of the device for detecting and recognizing the text of the automobile factory certificate provided by the embodiment of the present application;

[0071] Figure 13 The composition diagram of the electronic device provided by the embodiment of the present application. DETAILED DESCRIPTION

[0072] The present application will be further described below in conjunction with the embodiments and drawings. Obviously, the described embodiments are part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0073] Embodiment 1:

[0074] Reference Figure 1As shown, the text detection and recognition method for automobile factory qualified certificate provided by the embodiment mainly includes the following steps:

[0075] S1, acquire digital original images of a plurality of automobile factory qualified certificates as initial data sets;

[0076] S2, perform image preprocessing on each image in the initial data set to remove grid lines in the image and fade the seal in the image, and obtain a preprocessed image data set.

[0077] Specifically, referring to Figure 2 As shown, the removal of grid lines in the image includes the following sub-steps:

[0078] S201, convert the digital original image of the automobile factory qualified certificate into a grayscale image to facilitate subsequent computer processing and understanding of the image, reduce the complexity of the image data, reduce the amount of calculation, and thus improve the efficiency of image processing; then perform a global threshold operation on the grayscale image to obtain a binary image, to simplify the image processing process, eliminate noise interference in the original image, and improve the contrast and clarity of the image. This process can be represented by the following formula:

[0079]

[0080] In the formula, T represents the global threshold value, f(x, y) is the set threshold value, and g(x, y) is the changed threshold value;

[0081] S202, invert the pixel value of the binary image to facilitate the making of a mask and achieve extraction of the grid line area in the original grayscale image.

[0082] S203, analyze and extract the connected regions of the inverted binary image to obtain a grid line image, as shown in Figure 3 As shown, it specifically includes the following sub-steps:

[0083] S2031, in the horizontal or vertical direction of the inverted binary image, if adjacent pixel points have the same pixel value, they belong to the same connected region;

[0084] S2032, assign a unique label to each connected region;

[0085] S2033, based on the label, loop through all connected regions and calculate the number of pixels contained in the connected regions to determine the connected region with the most pixels. The connected region with the most pixels is the grid line, and the other connected regions may be strokes in the text.

[0086] S2034, the pixel of the most connected region is reserved, the remaining connected regions are removed, and a grid line image is obtained; thus, the pixel region corresponding to the grid line can be found, and the pixel region corresponding to the text is removed.

[0087] S204, the grid line image is dilated and eroded to remove noise points in the image, and a region mask is obtained.

[0088] S205, the gray-scale image is pixel inverted again, and mixed with the region mask at the pixel level.

[0089] Thus, the grid lines in the image can be removed by the above step operation, thereby avoiding the interference of the grid lines on the text detection and recognition.

[0090] Specifically, as shown in Figure 4 The image is faded, including:

[0091] S211, the digital original image of the automobile factory qualified certificate is globally thresholded and binarized;

[0092] S212, the digital original image of the automobile factory qualified certificate is fused with the globally binarized image based on weight; thus, the fusion based on weight can maintain the contrast and clarity of the text itself by accurately adjusting the fusion ratio (i.e. weight).

[0093] S213, the fused image is dilated and eroded.

[0094] Thus, the above step operation can fade the seal in the qualified certificate, reduce the influence of the seal on the text, and restore the clarity of the text covered by the seal.

[0095] S3, the preprocessed image dataset is divided into a training set and a test set, and the training set is labeled;

[0096] S4, an improved differentiable binarization network DBNet (Differentiable Binarization Network) is constructed, and the training set is used for training;

[0097] Specifically, in the main network ResNet18-vd (Residual Network, residual network), the first three ordinary 2D convolution kernels are replaced by deformable convolution kernels, which can be represented by the formula:

[0098]

[0099] In the formula, P0 represents any point on the input feature map, w represents the weight of the corresponding position of the convolution kernel, and P nrepresent the offset of each point in the convolution kernel relative to the center point, ΔP n represent the new offset;

[0100] In the feature fusion FPN (Feature Pyramid Network) part, two attention mechanisms, SENet (Squeeze-and-Excitation Networks) and CBAM (Convolutional Block Attention Module), are introduced.

[0101] In this way, by using a deformable convolution kernel instead of the original 2D convolution kernel, and introducing SENet and CBAM two attention mechanisms in the FPN structure, the accuracy and robustness of the convolution network for text detection and recognition are improved.

[0102] S5, inputting the test set into the improved differentiable binarization network to obtain a text detection and recognition result.

[0103] Therefore, the text detection and recognition method for automobile factory qualified certificates provided in the embodiment can not only reduce the influence of grid lines and seals on text detection, but also improve the accuracy of dense character recognition.

[0104] The text detection and recognition method for automobile factory qualified certificates provided in the present application will be further verified and described in combination with an application scenario example, as shown in Figure 5 , which includes:

[0105] Step one, 600 paper motor vehicle factory qualified certificates are obtained in the form of scanning to obtain digital images as initial data sets. The image size is 1260x1760, as shown in Figure 6 ;

[0106] Step two, pre-process all images, as shown in Figure 7 , 8 to obtain images after removing grid lines and fading seals;

[0107] Step three, divide the images obtained in step two into training set and test set according to the ratio of 5:1, and label. In the labeling process, distinguish the text that is close in space but irrelevant in content. The text box is mainly annotated horizontally, and for the inclined text caused by the inclination of the paper document, the arbitrary quadrilateral point-by-point annotation is used;

[0108] Step four, constructing the improved DBNet; because the number of data sets in step one is insufficient to complete network model training, in this example, the LSVT (Large-scale Street View Text with Partial Labeling) data set is used to pre-train the improved DBNet, and then the training set in step three is used for formal training;

[0109] Step five, inputting the test set in step three into the formally trained network to obtain text detection and recognition results.

[0110] Step six, comparing the text detection and recognition results obtained by using the improved DBNet after image preprocessing with the text detection and recognition results obtained by using the original DBNet without image preprocessing, as shown in Figures 9-11 It can be clearly seen that the accuracy of the detection using the method of the present application is higher, especially in the grid lines and the seal area, and the text can be accurately detected and recognized.

[0111] In summary, the text detection and recognition method for automobile factory qualified certificates provided in the embodiment has the following technical advantages compared with the prior art:

[0112] (1) The present application uses connected region analysis based on the traditional image processing method, which can remove the grid lines in the image, thereby avoiding the interference of the grid lines on the text detection and recognition;

[0113] (2) The present application proposes a method for fading the seal in the qualified certificate, which can reduce the influence of the seal on the text and restore the clarity of the text covered by the seal;

[0114] (3) The present application uses a deformable convolution kernel instead of the original 2D convolution kernel, and introduces SENet and CBAM two attention mechanisms in the FPN structure, which improves the accuracy and robustness of the convolution network for text detection and recognition, thereby reducing the probability of text missed detection and improving the accuracy of text detection and recognition.

[0115] Embodiment 2:

[0116] As shown in Figure 12 , the present embodiment provides a text detection and recognition device for automobile factory qualified certificates, which comprises:

[0117] A digital image acquisition module for acquiring digital original images of a plurality of automobile factory qualified certificates as an initial data set;

[0118] An image preprocessing module for performing image preprocessing on each image in the initial data set to remove the grid lines in the image and fade the seal in the image, thereby obtaining a preprocessed image data set;

[0119] A data set labeling module is configured to divide the preprocessed image data set into a training set and a test set, and label the training set;

[0120] An improved differentiable binarization network training module is configured to construct an improved differentiable binarization network, and train the improved differentiable binarization network using the training set.

[0121]

[0122] In the formula, P0 represents an arbitrary point on an input feature map, w represents a weight of a corresponding position of a convolution kernel, P n represents an offset of each point in the convolution kernel relative to the center point, and ΔP n represents a new offset.

[0123] In the feature fusion FPN (Feature Pyramid Network) part, two attention mechanisms, SENet (Squeeze-and-Excitation Networks) and CBAM (Convolutional Block Attention Module), are introduced.

[0124] A detection and recognition module is configured to input the test set into the improved differentiable binarization network to obtain a text detection and recognition result.

[0125] Specifically, the image preprocessing module includes a grid line removal submodule, and the grid line removal submodule includes:

[0126] A binarization image unit is configured to convert a digitized original image of a vehicle factory certificate into a gray-scale image, and perform a global threshold operation on the gray-scale image to obtain a binarization image.

[0127] A reverse unit is configured to perform pixel value inversion on the binarization image.

[0128] A grid line image unit is configured to analyze and extract a connected region of the inverted binarization image to obtain a grid line image.

[0129] A processing unit is configured to perform dilation and erosion processing on the grid line image to obtain a region mask.

[0130] A mixing unit is configured to perform pixel inversion on the gray-scale image again, and mix the pixel level with the region mask.

[0131] Specifically, the grid line image unit comprises:

[0132] a homing subunit, configured to, in the horizontal or vertical direction of the inverted binary image, if adjacent pixel points have the same pixel value, the pixel points are homed to the same connected region;

[0133] a label subunit, configured to assign a unique label to each connected region;

[0134] a determination subunit, configured to perform a loop traversal on all connected regions based on the label, and calculate the number of pixels contained in each connected region, and determine the connected region with the most pixels;

[0135] a reservation subunit, configured to reserve the connected region with the most pixels, and remove the remaining connected regions.

[0136] Specifically, the image preprocessing module further comprises a seal fading sub-module, and the seal fading sub-module comprises:

[0137] a global binaryzation unit, configured to perform global binaryzation on the digitized original image of the automobile factory qualified certificate;

[0138] a fusion unit, configured to fuse the digitized original image of the automobile factory qualified certificate and the globally binaryzated image based on a weight;

[0139] a processing unit, configured to perform dilation and corrosion processing on the fused image.

[0140] It should be noted that the text detection and recognition device for the automobile factory qualified certificate provided in the embodiments of the present application can execute the text detection and recognition method for the automobile factory qualified certificate provided in any embodiment of the present application, and has the corresponding functions and beneficial effects of the execution method.

[0141] In a specific implementation, the text detection and recognition device for the automobile factory qualified certificate can be integrated in a device, or can be a physical entity, for example, the electronic device can be a personal computer (PC), a computer, a server, etc., and the embodiments of the present application do not make a specific limitation thereto.

[0142] Embodiment 3

[0143] As Figure 13As shown, the embodiment of the present application provides an electronic device, comprising a processor 111, a communication interface 112, a memory 113 and a communication bus 114, wherein the processor 111, the communication interface 112 and the memory 113 complete mutual communication through the communication bus 114; the memory 113 is used for storing a computer program; the processor 111 is used for executing the program stored on the memory 113 to realize the steps of the text detection and recognition method for the automobile factory qualified certificate provided by any one of the method embodiments.

[0144] It should be noted that, in this document, relational terms such as "first" and "second", and the like, are used solely to distinguish one entity or action from another entity or action, without necessarily requiring or implying any actual relationship or order between such entities or actions. Moreover, the terms "comprises", "comprising", or any other variations thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but can include other elements not expressly listed or inherent to such process, method, article, or apparatus. An element proceeded by "comprises a... " does not, without more constraints, foreclose the existence of additional identical elements in the process, method, article, or apparatus that comprises the stated elements.

[0145] The above embodiments are only for illustrating the technical concept and characteristics of the present application, and the purpose is to enable those skilled in the art to understand the content of the present application and to implement it, and cannot limit the protection scope of the present application. Any equivalent changes or modifications made according to the essence of the present application should be covered within the protection scope of the present application.

Claims

1. A method for text detection and recognition of a vehicle certificate of conformity, characterized in that, The method comprises: acquiring a plurality of digital original images of vehicle factory qualification certificates as an initial data set; performing image preprocessing on each image in the initial data set to remove grid lines in the image and to fade a seal in the image, to obtain a preprocessed image data set; dividing the preprocessed image data set into a training set and a test set, and labeling the training set; constructing an improved differentiable binarization network and training the improved differentiable binarization network using the training set; inputting the test set into the improved differentiable binarization network to obtain a text detection and recognition result; the removing of the grid lines in the image comprises: converting the digital original image of the vehicle factory qualification certificate into a grayscale image, and performing a global threshold operation on the grayscale image to obtain a binarized image; inverting pixel values of the binarized image; performing analysis and extraction of connected regions on the inverted binarized image to obtain a grid line image; performing dilation and erosion processing on the grid line image to obtain a region mask; inverting the pixel values of the grayscale image again, and mixing the pixel values with the region mask at a pixel level; the converting of the digital image of the vehicle factory qualification certificate into a grayscale image and the performing of the global threshold operation on the grayscale image are represented as: where T represents a global threshold value, f(x, y) is a set threshold value, and g(x, y) is a changed threshold value; the improved differentiable binarization network is constructed in the following manner: in the backbone network ResNet18-vd, the first three 2D convolution kernels are replaced with deformable convolution kernels; in the feature fusion FPN part, two attention mechanisms, SENet and CBAM, are introduced; the deformable convolution kernel is represented as: In the formula, P0 represents an arbitrary point on an input feature map, w represents a weight of a corresponding position of a convolution kernel, P n represents an offset of each point in the convolution kernel relative to a center point, and ΔP n represents a new offset.

2. The method for text detection and recognition of a vehicle certificate of origin of claim 1, wherein, the analysis and extraction of connected regions on the inverted binarized image comprise: in the inverted binarized image, adjacent pixel points in the horizontal or vertical direction are attributed to the same connected region if they have the same pixel value; each connected region is assigned a unique label; all connected regions are traversed based on the label, and the number of pixels contained in each connected region is calculated to determine the connected region with the most pixels; the connected region with the most pixels is retained, and the remaining connected regions are removed.

3. The method for text detection and recognition of a vehicle certificate of origin of claim 1, wherein, the fading of the seal in the image comprises: globally binarizing the digital original image of the vehicle factory qualification certificate; fusing the digital original image of the vehicle factory qualification certificate with the globally binarized image based on a weight; performing dilation and erosion processing on the fused image.

4. A text detection and recognition apparatus for a vehicle certificate of origin, characterized by, The device comprises: a digital image acquisition module configured to acquire a plurality of digital original images of vehicle factory qualification certificates as an initial data set; an image preprocessing module configured to perform image preprocessing on each image in the initial data set to remove grid lines in the image and to fade a seal in the image, to obtain a preprocessed image data set; a data set labeling module configured to divide the preprocessed image data set into a training set and a test set, and to label the training set; an improved differentiable binarization network training module configured to construct an improved differentiable binarization network and to train the improved differentiable binarization network using the training set; and a text detection and recognition module configured to input the test set into the improved differentiable binarization network to obtain a text detection and recognition result. A detection and recognition module is configured to input the test set into the improved differentiable binarization network to obtain a text detection and recognition result. The image preprocessing module comprises a grid line removal submodule, which comprises: A binarization image unit is configured to convert a digital original image of the vehicle factory certificate into a gray-scale image, and perform a global threshold operation on the gray-scale image to obtain a binarization image. An inverse unit is configured to perform pixel value inversion on the binarization image. A grid line image unit is configured to analyze and extract a connected region of the inverted binarization image to obtain a grid line image. A processing unit is configured to perform dilation and erosion processing on the grid line image to obtain a region mask. A mixing unit is configured to perform pixel inversion on the gray-scale image again, and mix the pixel level with the region mask. The digital image of the vehicle factory certificate is converted into a gray-scale image, and a global threshold operation is performed on the gray-scale image, which is represented as: In the formula, T represents a global threshold, f(x, y) is a set threshold, and g(x, y) is a changed threshold. The improved differentiable binarization network is constructed in the following manner: In the backbone network ResNet18-vd, the first three 2D convolution kernels are replaced with deformable convolution kernels. In the feature fusion FPN part, two attention mechanisms, SENet and CBAM, are introduced. The deformable convolution kernel is represented as: In the formula, P0 represents an arbitrary point on an input feature map, w represents a weight of a corresponding position of a convolution kernel, P n represents an offset of each point in the convolution kernel relative to a center point, and ΔP n represents a new offset.

5. The text detection and recognition apparatus for a vehicle certificate of origin of claim 4, wherein, The image preprocessing module further comprises a seal fading submodule, which comprises: A global binarization unit is configured to perform global binarization on the digital original image of the vehicle factory certificate. A fusion unit is configured to fuse the digital original image of the vehicle factory certificate and the image after global binarization based on a weight. A processing unit is configured to perform dilation and erosion processing on the fused image.

6. An electronic device, comprising: The system comprises a processor, a communication interface, a memory and a communication bus, wherein the processor, the communication interface and the memory communicate with each other through the communication bus. The memory is configured to store a computer program. The processor is configured to execute the program stored on the memory to implement the steps of the text detection and recognition method for the vehicle factory certificate according to any one of claims 1-3.

Images