License plate recognition model training methods, license plate recognition methods, devices and related equipment

Abstract

This invention provides a license plate recognition model training method and a license plate recognition method. The method involves acquiring a license plate recognition model to be trained and sample license plate images; determining the fuzziness level of license plate characters within the sample license plate images based on the fuzziness levels; determining the corresponding license plate character labels based on the fuzziness levels of the license plate characters in the sample license plate images; and performing supervised training on the license plate recognition model to be trained based on the sample license plate images and the license plate character labels to obtain a trained license plate recognition model. By using the license plate recognition model obtained based on sample license plate images and corresponding fuzziness level license plate character labels to recognize target images, the method fully considers the loss relationship between label category probability and predicted category probability, thereby achieving accurate recognition of fuzzy license plates with high efficiency.

Description

Technical Field

[0001] This invention relates to the field of computer technology, and in particular to a license plate recognition model training method, license plate recognition method, device, electronic equipment, and storage medium. Background Technology

[0002] Driven by traffic safety management needs, the demand for license plate recognition in traffic images is increasing. Existing license plate recognition solutions typically employ hard labeling, aiming to make the character recognition probability vector and its label vector as close as possible. This method achieves high accuracy when the traffic image is clear. However, in blurry traffic images, license plate labeling is highly uncertain. Using hard labeling results in extremely low accuracy and a high risk of misjudgment. In such cases, manual identification of blurry license plates is often necessary, but this is inefficient and wastes resources. Therefore, providing a more efficient and accurate solution for recognizing blurry license plates is a pressing issue. Summary of the Invention

[0003] This invention provides a license plate recognition model training method and a license plate recognition method, aiming to solve the problems of low accuracy and low efficiency in existing fuzzy license plate recognition. The method determines the license plate character label corresponding to the license plate character in the sample license plate image by determining the fuzziness level of the characters. The license plate recognition model is then trained using the sample license plate image and the license plate character labels to obtain a trained license plate recognition model. Inputting the target image into the trained model yields the license plate recognition result. This method fully considers the loss relationship between the label category probability and the predicted category probability, thus achieving accurate and efficient recognition of fuzzy license plates.

[0004] In a first aspect, embodiments of the present invention provide a method for training a license plate recognition model, the method comprising:

[0005] Obtain the license plate recognition model to be trained and sample license plate images, wherein the sample license plate images include license plate characters;

[0006] The blur level of the license plate characters within the sample license plate image is determined based on the sample license plate image;

[0007] The license plate character label corresponding to the sample license plate image is determined based on the blur level of the license plate characters within the sample license plate image;

[0008] The license plate recognition model to be trained is subjected to supervised training based on the sample license plate images and the license plate character labels to obtain a trained license plate recognition model.

[0009] Optionally, determining the blur level of license plate characters within the sample license plate image based on the sample license plate image includes:

[0010] Based on the sample license plate image, determine the license plate character image block within the sample license plate image;

[0011] Based on the license plate character image block, determine the blur level of the license plate character image block;

[0012] Based on the blur level of the license plate character image block, the blur level of the license plate characters in the sample license plate image is determined.

[0013] Optionally, determining the license plate character label corresponding to the sample license plate image based on the blur level of the license plate characters within the sample license plate image includes:

[0014] The marker character of the license plate character in the sample license plate image is determined based on the blur level of the license plate character in the sample license plate image, and different blur levels correspond to different marker characters;

[0015] Based on the marker characters of the license plate characters in the sample license plate image, determine the license plate character label corresponding to the sample license plate image.

[0016] Optionally, the step of supervising the training of the license plate recognition model based on the sample license plate image and the license plate character labels to obtain a trained license plate recognition model includes:

[0017] Based on a preset character set and the license plate character labels, the label category probability of each license plate character in the sample license plate image is determined. The preset character set includes all categories of vehicle characters.

[0018] Based on the license plate recognition model to be trained, the predicted category probability of each license plate character in the sample license plate image is determined;

[0019] Based on the label category probability of each license plate character in the sample license plate image and the predicted category probability of each license plate character in the sample license plate image, the loss of each license plate character in the sample license plate image is determined.

[0020] The loss of each license plate character in the sample license plate image is optimized and trained to obtain a trained license plate recognition model.

[0021] Optionally, determining the label category probability of each license plate character in the sample license plate image based on a preset character set and the license plate character labels includes:

[0022] Determine the target character, which is any one of the marker characters in the license plate character label;

[0023] Based on the fuzziness level corresponding to the target character, a probability mapping strategy for the target character is determined, with different fuzziness levels corresponding to different probability mapping strategies;

[0024] Based on the probability mapping strategy of the target character, the tag category probability of the target character is determined in the preset character set.

[0025] Optionally, the fuzziness levels include a first fuzziness level, a second fuzziness level, and a normal level. The fuzziness of the first fuzziness level is greater than that of the second fuzziness level, and the fuzziness of the second fuzziness level is greater than that of the normal level. The step of determining the tag category probability of the target character in the preset character set based on the probability mapping strategy of the target character includes:

[0026] When the fuzziness level of the target character is the first fuzziness level, the first category probability of the target character is determined in the preset character set;

[0027] When the fuzziness level of the target character is the second fuzziness level, the index threshold corresponding to the target character is determined, and the second category probability of the target character is determined in the preset character set based on the index threshold;

[0028] When the target character is at the normal level, the probability of the third category of the target character is determined based on the character set;

[0029] The tag category probability of each target character is determined based on the first category probability, the second category probability, and the third category probability;

[0030] Based on the label category probability of each target character, the label category probability of each license plate character in the sample license plate image is determined.

[0031] Secondly, embodiments of the present invention also provide a license plate recognition method, the method comprising:

[0032] The target license plate image and the license plate recognition model are obtained, and the license plate recognition model is obtained through the above-described license plate recognition model training method;

[0033] The target license plate image is input into the license plate recognition model for recognition processing to obtain the license plate recognition result.

[0034] Thirdly, embodiments of the present invention also provide a license plate recognition model training device, the license plate recognition model training device comprising:

[0035] The first acquisition module is used to acquire the license plate recognition model to be trained and sample license plate images, wherein the sample license plate images include license plate characters;

[0036] The first determining module is used to determine the blur level of the license plate characters in the sample license plate image based on the sample license plate image;

[0037] The second determining module is used to determine the license plate character label corresponding to the sample license plate image based on the fuzziness level of the license plate characters;

[0038] The first training module is used to perform supervised training on the license plate recognition model to be trained based on the sample license plate image and the license plate character label, so as to obtain the trained license plate recognition model.

[0039] Fourthly, embodiments of the present invention also provide a license plate recognition device, the license plate recognition device comprising:

[0040] The second acquisition module is used to acquire the target license plate image and the license plate recognition model;

[0041] The first recognition module is used to input the target license plate image into the license plate recognition model for recognition processing to obtain the license plate recognition result.

[0042] Fifthly, embodiments of the present invention provide an electronic device, including: a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the steps in the license plate recognition model training method or license plate recognition method provided in the embodiments of the present invention.

[0043] In a sixth aspect, embodiments of the present invention provide a computer-readable storage medium storing a computer program, wherein when the computer program is executed by a processor, it implements the steps in the license plate recognition model training method or license plate recognition method provided in the embodiments of the present invention.

[0044] In this embodiment of the invention, a license plate recognition model to be trained and sample license plate images are obtained. The sample license plate images include license plate characters. The fuzziness level of the license plate characters in the sample license plate images is determined based on the fuzziness level. A license plate character label corresponding to the sample license plate characters is determined based on the fuzziness level. Supervised training is then performed on the license plate recognition model to be trained based on the sample license plate images and the license plate character labels, resulting in a trained license plate recognition model. By using the license plate recognition model obtained based on sample license plate images and corresponding fuzziness level license plate character labels to recognize target images, the loss relationship between the label category probability and the predicted category probability is fully considered, thereby achieving accurate recognition of fuzzy license plates with high recognition efficiency. Attached Figure Description

[0045] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0046] Figure 1 This is a flowchart of a license plate recognition model training method provided in an embodiment of the present invention;

[0047] Figure 2 This is a schematic diagram of a blurred license plate provided in an embodiment of the present invention;

[0048] Figure 3 This is a flowchart of a license plate recognition method provided in an embodiment of the present invention;

[0049] Figure 4 This is a schematic diagram of the structure of a license plate recognition model training device provided in an embodiment of the present invention;

[0050] Figure 5 This is a schematic diagram of the structure of a license plate recognition device provided in an embodiment of the present invention;

[0051] Figure 6 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0052] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0053] like Figure 1 As shown, Figure 1 This is a flowchart of a license plate recognition model training method provided by an embodiment of the present invention. The license plate recognition model training method includes the following steps:

[0054] 101. Obtain the license plate recognition model to be trained and sample license plate images.

[0055] In this embodiment of the invention, the license plate recognition model training method described above can be set on a server. The license plate recognition model trained on the server can be deployed within a traffic management platform. The traffic management platform can be any server or server cluster with functions such as image recognition, image processing, image storage, and data transmission. The license plate recognition model to be trained can be any neural network model with image processing and image recognition capabilities. The neural network model can be a convolutional neural network model, such as LeNet-5, AlexNet, or VGGNet. The sample license plate images can include license plate characters. The number of license plate characters is related to the license plate type; ordinary license plates have 7 characters, while new energy vehicle license plates have 8 characters. In the sample license plate images, the license plate characters can be blurred or clear.

[0056] In one possible embodiment, when it is necessary to train a license plate recognition model, the license plate recognition model to be trained and sample license plate images are loaded through the server.

[0057] 102. Determine the blur level of license plate characters in the sample license plate image based on the sample license plate image.

[0058] In this embodiment of the invention, the aforementioned blur level can be obtained by annotating the license plate image. The blur level can be determined based on the degree of blur of the license plate characters. The blur level of the license plate characters in the sample license plate image is determined based on their degree of blur. The degree of blur of the license plate characters in the sample license plate image corresponds to the blur level of the license plate characters in the sample license plate image. The degree of blur of multiple license plate characters in the sample license plate image corresponds to the blur level of multiple license plate characters in the sample license plate image. Specifically, the license plate characters can be classified according to their degree of blur into completely unrecognizable characters, characters with similar characters that are difficult to distinguish, and completely recognizable characters. The blur level of the license plate characters in the sample license plate image is determined based on these three categories: completely unrecognizable characters, characters with similar characters that are difficult to distinguish, and completely recognizable characters.

[0059] For example, it can be based on such Figure 2 The fuzzy license plate diagram shown is classified into fuzzy levels based on, for example, Figure 2 The degree of blurriness of the license plate characters in the illustrated blurred license plate diagram is determined. Figure 2 The ambiguity level of license plate characters, among which the above Figure 2In the license plate, the first character from left to right is completely unidentifiable. The second character from left to right is difficult to distinguish due to similar characters A or H. The third character from left to right is definitely 1. The fourth character from left to right is definitely L. The fifth character from left to right is difficult to distinguish due to similar characters 2 or Z. The sixth character from left to right is difficult to distinguish due to similar characters 8, 6, or B. The seventh character from left to right is definitely 9.

[0060] Specifically, the above Figure 2 The license plate contains a total of seven characters. The first character from left to right is completely unrecognizable; the second character is difficult to distinguish due to similar characters; the third character is completely identifiable; the fourth character is completely identifiable; the fifth character is difficult to distinguish due to similar characters; the sixth character is difficult to distinguish due to similar characters; and the seventh character is completely identifiable. Based on these degrees of ambiguity, it can be determined that... Figure 2 The first license plate character from left to right has the first level of blur. The second license plate character from left to right has the second level of blur. The third license plate character from left to right has the normal level of blur. The fourth license plate character from left to right has the normal level of blur. The fifth license plate character from left to right has the second level of blur. The sixth license plate character from left to right has the second level of blur. The seventh license plate character from left to right has the normal level of blur.

[0061] In one possible embodiment, after the server obtains the license plate recognition model to be trained and the sample license plate image, it determines the degree of blurring of the license plate characters based on the license plate characters in the sample license plate image, and determines the blur level of the license plate characters in the sample license plate image based on the degree of blurring.

[0062] In one possible embodiment, the aforementioned blur level can be obtained by automatically labeling the license plate image using methods such as Gradient Similarity (GSIM) and Detail Preservation (DP).

[0063] In one possible embodiment, after the server obtains the license plate recognition model to be trained and the sample license plate image, it determines the directional gradient similarity (GSIM) of the license plate characters in the sample license plate image based on the license plate characters in the sample license plate image, and determines the fuzziness level corresponding to the license plate characters in the sample license plate image based on the directional gradient similarity of the license plate characters in the sample license plate image.

[0064] Specifically, the aforementioned directional gradient similarity can be used to evaluate the blur level of the image by utilizing the gradient information of the sample license plate images. The lower the GSIM value corresponding to the license plate characters in the sample license plate images, the higher the blur level of the license plate characters in the sample license plate images. The higher the GSIM value corresponding to the license plate characters in the sample license plate images, the lower the blur level of the license plate characters in the sample license plate images. The higher the blur level of the license plate characters in the sample license plate images, the higher the blur level of the license plate characters in the sample license plate images. The lower the blur level of the license plate characters in the sample license plate images, the lower the blur level of the license plate characters in the sample license plate images.

[0065] In one possible embodiment, after the server obtains the license plate recognition model to be trained and the sample license plate image, it determines the detail preservation (DP) of the license plate characters in the sample license plate image based on the license plate characters in the sample license plate image, and determines the blur level corresponding to the license plate characters in the sample license plate image based on the detail preservation.

[0066] Specifically, the aforementioned detail retention can be used to assess the blurriness of the image based on the detail information of the sample license plate images. The lower the DP value corresponding to the license plate characters in the sample license plate images, the higher the blur level of the license plate characters in the sample license plate images. The higher the DP value corresponding to the license plate characters in the sample license plate images, the lower the blur level of the license plate characters in the sample license plate images. The higher the blur level of the license plate characters in the sample license plate images, the higher the degree of blurriness of the license plate characters in the sample license plate images. The lower the blur level of the license plate characters in the sample license plate images, the lower the degree of blurriness of the license plate characters in the sample license plate images.

[0067] 103. Determine the license plate character label corresponding to the sample license plate image based on the fuzziness level of the license plate characters in the sample license plate image.

[0068] In this embodiment of the invention, the license plate character label may include the blur level of the license plate character. The license plate character label may be used to specify the blur level of the license plate character in the sample license plate image. The sample license plate image includes multiple license plate characters, each license plate character corresponds to a blur level, and each blur level corresponds to a blur mark. The blur marks corresponding to the license plate characters in the sample license plate image are arranged to obtain the license plate character label corresponding to the sample license plate image. Each sample license plate image corresponds to one license plate character label, and the license plate character label includes the blur mark corresponding to each license plate character.

[0069] For example, if the number of characters corresponding to the license plate characters in the above sample license plate image is eight, then the blur level of the eight license plate characters in the above sample license plate image is determined. If four of the eight license plate characters are completely unrecognizable, three of the license plate characters are similar characters that are difficult to distinguish, and one of the license plate characters is completely recognizable, then based on the above blur level, the blur level corresponding to the four completely unrecognizable characters is determined as the first blur level, the blur level corresponding to the three similar characters that are difficult to distinguish is determined as the second blur level, and the blur level corresponding to the one completely recognizable character is determined as the third blur level. Based on the above first blur level, second blur level, and third blur level, corresponding license plate character labels are established to obtain the license plate character labels corresponding to the above sample license plate image.

[0070] In one possible embodiment, after determining the blur level of the license plate characters in the above sample license plate image, a corresponding license plate character label is established based on the blur level of the license plate characters in the above sample license plate image, and the license plate character label and the license plate character corresponding to the license plate character label are stored in the above traffic management platform.

[0071] In another possible embodiment, if the number of characters corresponding to the license plate characters in the above sample license plate image is seven, a blank license plate character can be set in the last position of the above license plate character label to fill it, and the above license plate character label and the license plate character corresponding to the above license plate character label can be stored in the above traffic management platform.

[0072] 104. Supervised training is performed on the license plate recognition model to be trained based on sample license plate images and license plate character labels to obtain a trained license plate recognition model.

[0073] In this embodiment of the invention, the supervised training described above can be a training method in model training. This involves inputting the sample license plate image and its corresponding license plate character label into the license plate recognition model to be trained for supervised training, thereby obtaining a trained license plate recognition model. Specifically, the sample license plate image can be input into the license plate recognition model to be trained for processing. Convolution processing is performed on the model to obtain intermediate features. Linear regression processing is then performed on the intermediate features through the output layer of the model to obtain the processing result corresponding to the sample license plate image. The error loss between the processing result and the license plate character label is calculated. Minimizing the error loss is used as the optimization objective. The parameters of the license plate recognition model to be trained are adjusted using a backpropagation algorithm. This parameter adjustment process is iterated until the error loss converges at its minimum, or the number of iterations reaches a preset value, at which point training stops, resulting in a trained license plate recognition model.

[0074] In this embodiment of the invention, a license plate recognition model to be trained and sample license plate images are obtained. The sample license plate images include license plate characters. The fuzziness level of the license plate characters in the sample license plate images is determined based on the fuzziness level. The corresponding license plate character label is determined based on the fuzziness level of the license plate characters in the sample license plate images. Supervised training is performed on the license plate recognition model to be trained based on the sample license plate images and the license plate character labels with corresponding fuzziness levels to obtain a trained license plate recognition model. By using the license plate recognition model obtained based on sample license plate images and license plate character labels with corresponding fuzziness levels to recognize the target image, the loss relationship between the label category probability and the predicted category probability is fully considered, thus achieving accurate recognition of fuzzy license plates with high efficiency.

[0075] Optionally, in the step of determining the blur level of license plate characters in the sample license plate image based on the sample license plate image, the following steps can be taken: determining license plate character image blocks of license plate characters in the sample license plate image based on the sample license plate image; determining the blur level of the license plate character image blocks based on the license plate character image blocks; and determining the blur level of the license plate characters in the sample license plate image based on the blur level of the license plate character image blocks.

[0076] In this embodiment of the invention, the above-mentioned sample license plate images can be segmented using an image segmentation algorithm, dividing the license plate characters within the sample vehicle images into corresponding license plate character image blocks, with each license plate character image block corresponding to one license plate character. The image segmentation algorithm can be the Roberts algorithm, Sobel algorithm, or Prewitt algorithm based on edge detection. Alternatively, it can calculate a grayscale threshold corresponding to the grayscale features of the above-mentioned sample license plate images, and perform image segmentation based on the grayscale threshold to obtain the corresponding license plate character image blocks.

[0077] Alternatively, the above-mentioned sample license plate image can be segmented using a preset image segmentation method. After segmentation, multiple license plate character image blocks are obtained. These multiple license plate character image blocks can be two or more, and each license plate character image block corresponds to a license plate character. Each license plate character image block contains one license plate character. The license plate character image blocks can be obtained by performing target detection on the above-mentioned sample license plate image to obtain the image coordinates of the license plate characters in the above-mentioned sample license plate image, and then dividing the region based on the image coordinates. The blur level of the license plate character image blocks can be determined by performing target detection on the above-mentioned license plate character image blocks to obtain the corresponding blur degree of the license plate character image blocks, determining the blur level of the license plate character image blocks based on the corresponding blur degree of the license plate character image blocks, and finally determining the blur level of the license plate characters in the above-mentioned sample license plate image based on the blur level of the license plate character image blocks.

[0078] In one possible embodiment, after the traffic management platform acquires the sample license plate image, it divides the sample license plate image into regions based on prior knowledge of license plates in the real world, obtaining license plate character image blocks corresponding to the license plate characters within the sample license plate image. The blur level of the license plate character image blocks is determined based on their blurriness, and the blur level of the license plate characters within the sample license plate image is determined based on the blur level of the license plate character image blocks. The prior knowledge of the license plate can be color prior or size prior. The color prior can be based on the background color and color of the license plate characters in the real world to determine the region corresponding to the license plate character image block. The size prior can be based on the size of the license plate characters in the real world, mapped to the size of the license plate characters in the sample license plate image to determine the region corresponding to the license plate character image block.

[0079] In one possible embodiment, after the traffic management platform acquires the sample license plate image, it obtains the image coordinates of the license plate characters corresponding to the license plate characters in the sample license plate image based on the color prior or the size prior. After dividing the region based on the image coordinates, a license plate character image block is obtained. The blur degree of the license plate character image block is determined based on the gradient similarity (GSIM) or detail preservation (DP) of the license plate character image block. The blur level of the license plate character image block is determined based on the blur degree of the license plate character image block. Finally, the blur level of the license plate characters in the sample license plate image is determined based on the blur level of the license plate character image block.

[0080] Optionally, in the step of determining the license plate character label corresponding to the sample license plate image based on the blur level of the license plate characters in the sample license plate image, the marker character of the license plate characters in the sample license plate image can be determined based on the blur level of the license plate characters in the sample license plate image, with different blur levels corresponding to different marker characters; and the license plate character label corresponding to the sample license plate image is determined based on the marker character of the license plate characters in the sample license plate image.

[0081] In this embodiment of the invention, the aforementioned marker character may be used to mark the blur level of the license plate characters in the aforementioned sample license plate image. The license plate characters in the aforementioned sample license plate image are marked based on the blur level of the license plate characters in the aforementioned sample license plate image to obtain the marker character of the license plate characters in the aforementioned sample license plate image. Different blur levels correspond to different marker characters, and the license plate character label corresponding to the aforementioned sample license plate image is determined based on the aforementioned marker character.

[0082] For example, if such as Figure 2 The blurred license plate diagram shown is used as the sample license plate image mentioned above. Therefore, based on the above... Figure 2 The fuzziness level of the license plate characters is marked with corresponding characters.

[0083] Specifically, the marker character corresponding to license plate characters with a fuzziness level of first fuzziness can be determined as "*", the marker character corresponding to license plate characters with a fuzziness level of second fuzziness can be determined as possible characters and "?", and the marker character corresponding to license plate characters with a fuzziness level of normal fuzziness can be determined as the character itself. Figure 2The first license plate character from left to right has a blur level of 1, so its marker character is "*". The second license plate character from left to right has a blur level of 2, so its marker character is "A?". The third license plate character from left to right has a normal blur level, so its marker character is "1". The fourth license plate character from left to right... The corresponding blur level is normal blur level. Therefore, the marker character for the fourth license plate character from left to right is "L". The blur level for the fifth license plate character from left to right is second blur level. Therefore, the marker character for the fifth license plate character from left to right is "2?". The blur level for the sixth license plate character from left to right is second blur level. Therefore, the marker character for the sixth license plate character from left to right is "8?". The blur level for the seventh license plate character from left to right is normal blur level. Therefore, the marker character for the seventh license plate character from left to right is "9". In summary, if... Figure 2 The blurred license plate diagram shown is used as the sample license plate image above, which can then determine the above... Figure 2 The fuzziness level of the license plate characters corresponds to the license plate character labels [*, A?, 1, L, 2?, 8?, 9]. If the maximum number of digits in the above license plate characters is eight, then the above... Figure 2 Since Chinese license plates have seven characters, which is less than eight, the symbol "#" can be used to represent a space character. Figure 2 The license plate character labels corresponding to the fuzziness levels of the license plate characters are [*, A?, 1, L, 2?, 8?, 9, #].

[0084] Optionally, in the step of supervising the training of the license plate recognition model to be trained based on sample license plate images and license plate character labels to obtain a trained license plate recognition model, the following steps can be taken: First, determine the label category probability of each license plate character in the sample license plate image based on a preset character set and license plate character labels. The preset character set includes all categories of vehicle characters. Second, determine the predicted category probability of each license plate character in the sample license plate image based on the license plate recognition model to be trained. Third, determine the loss of each license plate character in the sample license plate image based on the label category probability and the predicted category probability of each license plate character in the sample license plate image. Fourth, optimize and train the loss of each license plate character in the sample license plate image to obtain a trained license plate recognition model.

[0085] In this embodiment of the invention, the preset character set may include all categories of vehicle characters, which may include vehicle characters of the numeric category, vehicle characters of the English alphabet category, and vehicle characters of the Chinese character category. The label category probability may be the probability that each license plate character is located in the preset character set. The predicted category probability may be obtained by performing softmax calculation on the output layer of the license plate recognition model to be trained, calculating the loss between the predicted category probability and the label category probability through a loss function, and adjusting the parameters of the license plate recognition model to be trained based on the loss of each license plate character in the sample license plate image.

[0086] Specifically, the predicted category probability can be calculated using the following formula:

[0087]

[0088] Among them, the above P i,j,c For the predicted category probabilities mentioned above, 'i' represents the i-th sample license plate image, 'j' represents the j-th license plate character, 'c' represents the c-th character category, 'F' is the output of the output layer, and 'exp' is an exponential function with base e. The above P... i,j,c Specifically, it represents the probability that the j-th license plate character in the i-th sample license plate image belongs to the c-th character category.

[0089] The predicted category probabilities can also be obtained by inputting a preset character set into the license plate recognition model to be trained, obtaining each predicted category corresponding to the preset character set, and drawing a mapping table based on the predicted categories to obtain the predicted category probabilities.

[0090] Specifically, the loss function described above can be defined by the following formula:

[0091]

[0092] Where L is the loss function, N represents the batch size of the output layer of the license plate recognition model to be trained, K represents the maximum number of characters in the license plate characters output by the output layer of the license plate recognition model to be trained, i represents the number of sample license plates input to the license plate recognition model to be trained, j represents the number of license plate characters input to the license plate recognition model to be trained, c represents the probability that the number of license plate characters input to the license plate recognition model to be trained belongs to the c-th character category, and P i,j,c For the predicted category probabilities mentioned above, Q i,j,c For the above label category probabilities, specifically, the above P i,j,c Specifically, Q represents the probability that the j-th license plate character in the i-th sample license plate image belongs to the c-th character category.i,j,c Specifically, it represents the probability that the j-th license plate character label in the label corresponding to the i-th sample license plate image belongs to the c-th character label category.

[0093] Optionally, in the step of determining the label category probability of each license plate character in the sample license plate image based on a preset character set and license plate character labels, a target character can be determined, which is any one of the labeled characters in the license plate character labels; based on the fuzziness level corresponding to the target character, a probability mapping strategy for the target character is determined, with different fuzziness levels corresponding to different probability mapping strategies; based on the probability mapping strategy of the target character, the label category probability of the target character is determined in the preset character set.

[0094] In this embodiment of the invention, the target character can be any marked character in the sample license plate image. The different blur levels can include a first blur level, a second blur level, and a normal level. The first blur level can correspond to characters that are completely unrecognizable, the second blur level can correspond to characters that are difficult to distinguish due to similarity, and the normal blur level can correspond to characters that are completely identifiable. The probability mapping strategy can be as follows: when the blur level of the target character is the first blur level, determine the first category probability of the target character in a preset character set; when the blur level of the target character is the second blur level, determine the index corresponding to the target character. A threshold is set, and the second category probability of the target character is determined in a preset character set based on the index threshold; when the target character is of normal level, the third category probability of the target character is determined according to the character set; the label category probability of each target character is determined based on the first category probability, the second category probability, and the third category probability; based on the label category probability of each target character, or the fuzzy level corresponding to the above target character can be mapped to the above mapping table to obtain a probability mapping table, the above probability mapping strategy is determined based on the probability distribution of the above probability mapping table, and the label category probability of the above target character is determined in the preset character set based on the above probability mapping strategy.

[0095] Optionally, the fuzziness levels include a first fuzziness level, a second fuzziness level, and a normal level. The fuzziness of the first fuzziness level is greater than that of the second fuzziness level, and the fuzziness of the second fuzziness level is greater than that of the normal level. In the step of determining the label category probability of a target character in a preset character set based on the probability mapping strategy of the target character, when the fuzziness level of the target character is the first fuzziness level, the first category probability of the target character can be determined in the preset character set; when the fuzziness level of the target character is the second fuzziness level, the index threshold corresponding to the target character is determined, and the second category probability of the target character is determined in the preset character set based on the index threshold; when the target character is at the normal level, the third category probability of the target character is determined according to the character set; the label category probability of each target character is determined based on the first category probability, the second category probability, and the third category probability; and the label category probability of each license plate character in the sample license plate image is determined based on the label category probability of each target character.

[0096] In this embodiment of the invention, if the fuzziness level of the target character is the first fuzziness level, the total number of characters in the preset character set can be determined to be M based on the preset character set, and the first category probability can be determined based on M. Right now In the above example, q represents the category probability, and m represents the target character. The total number M of characters in the aforementioned preset character set can include M1, M2, and M3. M1 can be the number of Chinese characters, M2 can be the number of English letters, and M3 can be the number of numeric characters. If the fuzziness level of the target character is normal, then the corresponding index B is determined based on the target character with the normal fuzziness level. When the target character equals the index B, the probability of the third category is 1; when the target character does not equal the index B, the probability of the third category is 0. If the fuzziness level of the target character is second fuzziness level, then the total number M of characters in the aforementioned preset character set can be determined based on the aforementioned preset character set. The corresponding index B is determined based on the target character with the second fuzziness level, and a corresponding index threshold is set based on the index B. The index threshold thr can be a number less than 1, such as 0.5. When the target character equals the index B, the probability of the second category is equal to the index threshold thr. When the target character does not equal the index B, the probability of the second category can be expressed by the following formula:

[0097]

[0098] Where q represents the category probability, m represents the target character, and thr represents the index threshold.

[0099] In one possible embodiment, when the license plate character is the first character, if the fuzziness level of the first character is level 1, the probability of the first category is 1 / M1; if the fuzziness level of the first character is level 2, the probability of the second category is 1-thr / M1-1. When the license plate character is the second character, if the fuzziness level of the second character is level 1, the probability of the first category is 1 / M2; if the fuzziness level of the second character is level 2, the probability of the second category is 1-thr / M2-1. The second character is the character following the first character, and the first character is the first character in the license plate character set. When the license plate character is the last character, if the fuzziness level of the last character is level 1, the probability of the first category is 1 / M2+M3; if the fuzziness level of the last character is level 2, the probability of the second category is 1-thr / (M2+M3)-1. The last character is the character following the second character.

[0100] In one possible embodiment, after determining the fuzziness level of the target character, when the fuzziness level of the target character is the first fuzziness level, the first category probability corresponding to the target character is determined according to the above probability mapping strategy; when the fuzziness level of the target character is the second fuzziness level, the second category probability corresponding to the target character is determined according to the above probability mapping strategy; when the fuzziness level of the target character is the normal level, the third category probability corresponding to the target character is determined according to the above probability mapping strategy; the label category probability of the target character is determined according to the first category probability, the second category probability, and the third category probability; and the label category probability of the license plate character in the above sample license plate image is determined based on the label category probability of the target character.

[0101] It should be noted that the license plate recognition model training method provided in this embodiment of the invention can be applied to devices such as shooting devices, smartphones, computers, and servers that can train license plate recognition models.

[0102] like Figure 3 As shown, Figure 3 This is a flowchart of a license plate recognition method provided in an embodiment of the present invention. The license plate recognition method includes the following steps:

[0103] 301. Obtain the target license plate image and license plate recognition model.

[0104] In this embodiment of the invention, the above-mentioned license plate recognition method can be deployed in a license plate recognition platform. The license plate recognition platform can be a server or server cluster with functions such as image recognition, image processing, image storage, and data transmission. The target license plate image can be an image that needs to be recognized. The target license plate image can be an image captured by electronic devices such as road cameras, cameras, and mobile phone cameras. The license plate recognition model can be obtained by training a license plate recognition model to be trained using the above-mentioned license plate recognition model training method.

[0105] In one possible embodiment, after the license plate recognition platform receives the license plate recognition instruction, it acquires the target license plate image and the license plate recognition model through the data transmission function.

[0106] 302. Input the target license plate image into the license plate recognition model for recognition processing to obtain the license plate recognition result.

[0107] In this embodiment of the invention, the above recognition process can be performed by the above license plate recognition model, and the above license plate recognition result can be a recognition character set obtained after recognition processing based on the above license plate recognition model. The characters in the above recognition character set correspond to the characters in the above target license plate image.

[0108] In one possible embodiment, after the license plate recognition platform obtains the target license plate image and the license plate recognition model, it inputs the target license plate image into the license plate recognition model for recognition processing to obtain the license plate recognition result.

[0109] In this embodiment of the invention, a license plate recognition model to be trained and sample license plate images are obtained. The sample license plate images include license plate characters. The fuzziness level of the license plate characters in the sample license plate images is determined based on the fuzziness level. A license plate character label corresponding to the sample license plate characters is determined based on the fuzziness level. Supervised training is performed on the license plate recognition model to be trained based on the sample license plate images and the license plate character labels to obtain a trained license plate recognition model. A target license plate image is obtained. The target license plate image is input into the trained license plate recognition model for recognition processing to obtain the license plate recognition result. By using a license plate recognition model obtained based on sample license plate images and license plate character labels with corresponding fuzziness levels to perform license plate recognition on target images, the loss relationship between label category probability and predicted category probability is fully considered, thereby achieving accurate recognition of fuzzy license plates with high recognition efficiency.

[0110] It should be noted that the license plate recognition method provided in this embodiment of the invention can be applied to devices such as camera devices, smartphones, computers, and servers that can perform license plate recognition.

[0111] like Figure 4As shown, this embodiment of the invention provides a license plate recognition model training device, which includes:

[0112] The first acquisition module 401 is used to acquire the license plate recognition model to be trained and sample license plate images, wherein the sample license plate images include license plate characters;

[0113] The first determining module 402 is used to determine the blur level of the license plate characters in the sample license plate image based on the sample license plate image;

[0114] The second determining module 403 is used to determine the license plate character label corresponding to the sample license plate image based on the fuzziness level of the license plate characters;

[0115] The first training module 404 is used to perform supervised training on the license plate recognition model to be trained based on the sample license plate image and the license plate character label, so as to obtain a trained license plate recognition model.

[0116] Optionally, the first determining module 402 includes:

[0117] The first determining submodule is used to determine the license plate character image block of the license plate characters in the sample license plate image based on the sample license plate image;

[0118] The second determining submodule is used to determine the blur level of the license plate character image block based on the license plate character image block;

[0119] The third determining submodule is used to determine the blur level of the license plate characters in the sample license plate image based on the blur level of the license plate character image block.

[0120] Optionally, the second determining module 403 includes:

[0121] The fourth determination submodule is used to determine the marker character of the license plate character in the sample license plate image based on the blur level of the license plate character in the sample license plate image. Different blur levels correspond to different marker characters.

[0122] The fifth determination submodule is used to determine the license plate character label corresponding to the sample license plate image based on the marker character of the license plate character in the sample license plate image.

[0123] Optionally, the first training module 404 includes:

[0124] The sixth determination submodule is used to determine the label category probability of each license plate character in the sample license plate image based on a preset character set and the license plate character labels. The preset character set includes all categories of vehicle characters.

[0125] The seventh determination submodule is used to determine the predicted category probability of each license plate character in the sample license plate image based on the license plate recognition model to be trained.

[0126] The eighth determination submodule is used to determine the loss function of each license plate character in the sample license plate image based on the label category probability of each license plate character in the sample license plate image and the predicted category probability of each license plate character in the sample license plate image.

[0127] The first training submodule is used to optimize and train the loss function of each license plate character in the sample license plate image to obtain a trained license plate recognition model.

[0128] Optionally, the sixth determining submodule includes:

[0129] The first determining unit is used to determine the target character, wherein the target character is any one of the marker characters in the license plate character label;

[0130] The second determining unit is used to determine the probability mapping strategy of the target character based on the fuzziness level corresponding to the target character, wherein different fuzziness levels correspond to different probability mapping strategies;

[0131] The third determining unit is used to determine the tag category probability of the target character in the preset character set based on the probability mapping strategy of the target character.

[0132] Optionally, the third determining unit includes:

[0133] The first determining subunit is used to determine the first category probability of the target character in the preset character set when the fuzziness level of the target character is the first fuzziness level.

[0134] The second determining subunit is used to determine the index threshold corresponding to the target character when the fuzziness level of the target character is the second fuzziness level, and to determine the second category probability of the target character in the preset character set based on the index threshold.

[0135] The third determining subunit is used to determine the third category probability of the target character based on the character set when the target character is the normal level.

[0136] The fourth determining subunit is used to determine the tag category probability of each target character based on the first category probability, the second category probability, and the third category probability;

[0137] The fifth determining subunit is used to determine the label category probability of each license plate character in the sample license plate image based on the label category probability of each target character.

[0138] It should be noted that the license plate recognition model training device provided in this embodiment of the invention can be applied to devices such as shooting equipment, smartphones, computers, and servers that can train license plate recognition models.

[0139] The license plate recognition model training device provided in this embodiment of the invention can implement all the processes of the license plate recognition model training method in the above-described method embodiments, and can achieve the same beneficial effects. To avoid repetition, it will not be described again here.

[0140] like Figure 5 As shown, an embodiment of the present invention provides a license plate recognition device, which includes:

[0141] The second acquisition module is used to acquire the target license plate image and the license plate recognition model;

[0142] The first recognition module is used to input the target license plate image into the license plate recognition model for recognition processing to obtain the license plate recognition result.

[0143] It should be noted that the license plate recognition device provided in this embodiment of the invention can be applied to devices such as camera devices, smartphones, computers, and servers that can perform license plate recognition.

[0144] The license plate recognition device provided in this embodiment of the invention can implement all the processes of the license plate recognition method in the above-described method embodiments, and can achieve the same beneficial effects. To avoid repetition, it will not be described again here.

[0145] See Figure 6 , Figure 6 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention, such as... Figure 6 As shown, it includes: a memory 602, a processor 601, and a computer program for a license plate recognition method stored in the memory 602 and executable on the processor 601, wherein:

[0146] The processor 601 is used to call the computer program stored in the memory 602 and perform the following steps:

[0147] Obtain the license plate recognition model to be trained and sample license plate images, wherein the sample license plate images include license plate characters;

[0148] The blur level of the license plate characters within the sample license plate image is determined based on the sample license plate image;

[0149] The license plate character label corresponding to the sample license plate image is determined based on the blur level of the license plate characters within the sample license plate image;

[0150] The license plate recognition model to be trained is subjected to supervised training based on the sample license plate images and the license plate character labels to obtain a trained license plate recognition model.

[0151] Optionally, the step of determining the blur level of license plate characters within the sample license plate image based on the sample license plate image, executed by the processor 601, includes:

[0152] Based on the sample license plate image, determine the license plate character image block within the sample license plate image;

[0153] Based on the license plate character image block, determine the blur level of the license plate character image block;

[0154] Based on the blur level of the license plate character image block, the blur level of the license plate characters in the sample license plate image is determined.

[0155] Optionally, the process executed by processor 601 to determine the license plate character label corresponding to the sample license plate image based on the blur level of the license plate characters within the sample license plate image includes:

[0156] The marker character of the license plate character in the sample license plate image is determined based on the blur level of the license plate character in the sample license plate image, and different blur levels correspond to different marker characters;

[0157] Based on the marker characters of the license plate characters in the sample license plate image, determine the license plate character label corresponding to the sample license plate image.

[0158] Optionally, the process executed by processor 601 to supervise the training of the license plate recognition model to be trained based on the sample license plate image and the license plate character label to obtain a trained license plate recognition model includes:

[0159] Based on a preset character set and the license plate character labels, the label category probability of each license plate character in the sample license plate image is determined. The preset character set includes all categories of vehicle characters.

[0160] Based on the license plate recognition model to be trained, the predicted category probability of each license plate character in the sample license plate image is determined;

[0161] Based on the label category probability of each license plate character in the sample license plate image and the predicted category probability of each license plate character in the sample license plate image, the loss function of each license plate character in the sample license plate image is determined.

[0162] The loss function of each license plate character in the sample license plate image is optimized and trained to obtain a trained license plate recognition model.

[0163] Optionally, the processor 601 executes the process of determining the label category probability of each license plate character in the sample license plate image based on a preset character set and the license plate character labels, including:

[0164] Determine the target character, which is any one of the marker characters in the license plate character label;

[0165] Based on the fuzziness level corresponding to the target character, a probability mapping strategy for the target character is determined, with different fuzziness levels corresponding to different probability mapping strategies;

[0166] Based on the probability mapping strategy of the target character, the tag category probability of the target character is determined in the preset character set.

[0167] Optionally, the probability mapping strategy based on the target character executed by the processor 601, which determines the tag category probability of the target character in the preset character set, includes:

[0168] When the fuzziness level of the target character is the first fuzziness level, the first category probability of the target character is determined in the preset character set;

[0169] When the fuzziness level of the target character is the second fuzziness level, the index threshold corresponding to the target character is determined, and the second category probability of the target character is determined in the preset character set based on the index threshold;

[0170] When the target character is at the normal level, the probability of the third category of the target character is determined based on the character set;

[0171] The tag category probability of each target character is determined based on the first category probability, the second category probability, and the third category probability;

[0172] Based on the label category probability of each target character, the label category probability of each license plate character in the sample license plate image is determined.

[0173] The processor 601 is used to call computer programs stored in the memory 602, and can also perform the following steps:

[0174] The target license plate image and the license plate recognition model are obtained, wherein the license plate recognition model is obtained by any of the license plate recognition model training methods as described in claims 1-6;

[0175] The target license plate image is input into the license plate recognition model for recognition processing to obtain the license plate recognition result.

[0176] It should be noted that the electronic device provided in the embodiments of the present invention can be applied to devices such as smartphones, computers, and servers that can perform license plate recognition model training methods or license plate recognition methods.

[0177] The electronic device provided in this embodiment of the invention can implement all the processes of the license plate recognition model training method or the license plate recognition method in the above-described method embodiments, and can achieve the same beneficial effects. To avoid repetition, further details are omitted here.

[0178] This invention also provides a computer-readable storage medium storing a computer program. When the computer program is executed by a processor, it implements the various processes of the license plate recognition model training method, license plate recognition method, or application-side license plate recognition model training method and license plate recognition method provided in this invention, and achieves the same technical effect. To avoid repetition, it will not be described again here.

[0179] Those skilled in the art will understand that implementing all or part of the processes in the above embodiments can be accomplished by a computer program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. The storage medium can be a magnetic disk, optical disk, read-only memory (ROM), or random access memory (RAM), etc.

[0180] The above description discloses only preferred embodiments of the present invention and should not be construed as limiting the scope of the present invention. Therefore, equivalent variations made in accordance with the claims of the present invention are still within the scope of the present invention.

Claims

1. A method for training a license plate recognition model, the method comprising: The method comprises the following steps: Obtain a license plate recognition model to be trained and a sample license plate image, the sample license plate image comprising license plate characters; Determine the blur level of the license plate characters in the sample license plate image based on the sample license plate image; Determine the license plate character label corresponding to the sample license plate image based on the blur level of the license plate characters in the sample license plate image; Supervise the training of the license plate recognition model to be trained based on the sample license plate image and the license plate character label, and obtain a trained license plate recognition model; The supervision training of the license plate recognition model to be trained based on the sample license plate image and the license plate character label to obtain a trained license plate recognition model comprises: Determine the label category probability of each license plate character in the sample license plate image based on a preset character set and the license plate character label, the preset character set comprising all categories of vehicle characters; Determine the predicted category probability of each license plate character in the sample license plate image based on the license plate recognition model to be trained; Determine the loss of each license plate character in the sample license plate image based on the label category probability of each license plate character in the sample license plate image and the predicted category probability of each license plate character in the sample license plate image; Optimize the training of the loss of each license plate character in the sample license plate image to obtain a trained license plate recognition model; The determination of the label category probability of each license plate character in the sample license plate image based on a preset character set and the license plate character label comprises: Determine a target character, the target character being any one of the marked characters in the license plate character label; Determine the probability mapping strategy of the target character based on the blur level corresponding to the target character, different blur levels corresponding to different probability mapping strategies; Determine the label category probability of the target character in the preset character set based on the probability mapping strategy of the target character; The blur level comprises a first blur level, a second blur level, and a normal level, the blur degree of the first blur level being greater than the blur degree of the second blur level, the blur degree of the second blur level being greater than the blur degree of the normal level, and the determination of the label category probability of the target character in the preset character set based on the probability mapping strategy of the target character comprising: When the blur level of the target character is the first blur level, determine the first category probability of the target character in the preset character set; When the blur level of the target character is the second blur level, determine the index threshold corresponding to the target character, and determine the second category probability of the target character in the preset character set based on the index threshold; When the target character is the normal level, determine the third category probability of the target character according to the character set; Determine the label category probability of each target character based on the first category probability, the second category probability, and the third category probability. Determine a label category probability of each character in the sample license plate image based on the label category probability of each target character.

2. The vehicle license plate recognition model training method of claim 1, wherein, The determination of the blur level of the license plate character in the sample license plate image based on the sample license plate image comprises: Determine a license plate character image block of the license plate character in the sample license plate image based on the sample license plate image; Determine a blur level of the license plate character image block based on the license plate character image block; Determine the blur level of the license plate character in the sample license plate image based on the blur level of the license plate character image block. 3.The license plate recognition model training method of claim 1, wherein, The determination of the license plate character label corresponding to the sample license plate image based on the blur level of the license plate character in the sample license plate image comprises: Determine a marked character of the license plate character in the sample license plate image based on the blur level of the license plate character in the sample license plate image, different blur levels corresponding to different marked characters; Determine the license plate character label corresponding to the sample license plate image based on the marked character of the license plate character in the sample license plate image.

4. A license plate recognition method characterized by, The method comprises the following steps: Obtain a target license plate image and a license plate recognition model, wherein the license plate recognition model is obtained by using any one of the license plate recognition model training methods in claims 1-3; Input the target license plate image into the license plate recognition model for recognition processing to obtain a license plate recognition result. 5.A license plate recognition model training apparatus, characterized in that, The license plate recognition model training device comprises: A first obtaining module is configured to obtain a license plate recognition model to be trained and a sample license plate image, wherein the sample license plate image comprises a license plate character; A first determining module is configured to determine a blur level of the license plate character in the sample license plate image based on the sample license plate image; A second determining module is configured to determine a license plate character label corresponding to the sample license plate image based on the blur level of the license plate character; A first training module is configured to perform supervised training on the license plate recognition model to be trained based on the sample license plate image and the license plate character label to obtain a trained license plate recognition model; The supervised training on the license plate recognition model to be trained based on the sample license plate image and the license plate character label to obtain a trained license plate recognition model comprises: Determine a label category probability of each character in the sample license plate image based on a preset character set and the license plate character label, wherein the preset character set comprises all types of vehicle characters; Determine a predicted category probability of each character in the sample license plate image based on the license plate recognition model to be trained; Determine a loss of each character in the sample license plate image based on the label category probability of each character in the sample license plate image and the predicted category probability of each character in the sample license plate image; Optimize the loss of each character in the sample license plate image to obtain a trained license plate recognition model; The determination of the label category probability of each character in the sample license plate image based on the preset character set and the license plate character label comprises: Determine a target character, wherein the target character is any one of the marked characters in the license plate character label; determine a probability mapping strategy of the target character based on the blur level corresponding to the target character, different blur levels corresponding to different probability mapping strategies; determine a label category probability of the target character in the preset character set based on the probability mapping strategy of the target character; The blur level includes a first blur level, a second blur level, and a normal level, the blur degree of the first blur level is greater than the blur degree of the second blur level, the blur degree of the second blur level is greater than the blur degree of the normal level, and the determination of the label category probability of the target character in the preset character set based on the probability mapping strategy of the target character includes: When the blur level of the target character is the first blur level, a first category probability of the target character is determined in the preset character set; When the blur level of the target character is the second blur level, an index threshold corresponding to the target character is determined, and a second category probability of the target character is determined in the preset character set based on the index threshold; When the target character is the normal level, a third category probability of the target character is determined according to the character set; The label category probability of each target character is determined based on the first category probability, the second category probability, and the third category probability; The label category probability of each character in the sample license plate image is determined based on the label category probability of each target character.

6. A license plate recognition apparatus characterized by comprising: The license plate recognition device includes: The second acquisition module is configured to acquire a target license plate image and the license plate recognition model, and the license plate recognition model is obtained by the license plate recognition model training method in any one of claims 1-3. The first recognition module is configured to input the target license plate image into the license plate recognition model for recognition processing to obtain a license plate recognition result.

7. An electronic device, comprising: The license plate recognition device includes: A memory, a processor, and a computer program stored on the memory and executable on the processor, wherein the processor executes the computer program to implement the steps in the license plate recognition model training method in any one of claims 1-3 or the steps in the license plate recognition method in claim 4.

8. A computer-readable storage medium, characterized in that, The computer program is stored on the computer readable storage medium, and the computer program is executed by the processor to implement the steps in the license plate recognition model training method in any one of claims 1-3 or the steps in the license plate recognition method in claim 4.

Images