Image text recognition method and device for image scene

Abstract

The invention relates to the technical field of image text recognition, and discloses an image text recognition method and device for an image scene, and the method comprises the steps: obtaining a to-be-recognized text image; graying, filtering, denoising and layering the to-be-recognized text image, and dividing the to-be-recognized text image into a text layer and a background layer; segmenting the text layer to obtain a plurality of text region frames; judging the text in the text area frame, and determining the text type in the text area frame, including a handwritten text and a printed text; when the text type is a printed text, text characters are obtained according to a standard text recognition model; and when the text type is a handwritten text, determining the stroke trend of the text in the text region frame, and determining text characters according to the stroke trend. According to the invention, through combining image preprocessing, text layering, text type determination and a targeted text identification strategy, efficient and accurate image text identification can be realized in a complex scene.

Description

Technical Field

[0001] The present invention relates to the technical field of image text recognition, and in particular to an image text recognition method and device for image scenes. Background Art

[0002] In the field of image text recognition technology, with the rapid development of information technology and the popularization of smart devices, image text recognition has become an important research direction in the field of computer vision. Traditional image text recognition methods are mostly aimed at printed text and rely on high-quality image input. However, in practical applications, image text often comes from complex and changeable scenes, such as outdoor billboards, book covers, product packaging, etc. These images not only contain rich background information, but also may have problems such as uneven lighting, blur, and occlusion, which brings great challenges to image text recognition. Therefore, there is an urgent need for an image text recognition method and device for image scenes to reduce the interference of image scenes and improve the accuracy and robustness of image text recognition. Summary of the invention

[0003] The purpose of the present invention is to provide an image text recognition method and device for image scenes, so as to solve the problem of interference of image scenes on text recognition.

[0004] The present invention provides an image text recognition method for an image scene, the method comprising:

[0005] Get the text image to be recognized;

[0006] grayscale processing and filtering and denoising processing are performed on the text image to be recognized, and the processed text image to be recognized is layered into a text layer and a background layer;

[0007] Segmenting the text layer to obtain a number of text area frames;

[0008] Judging the text in the text area box to determine the text type in the text area box, the text type including handwritten text and printed text;

[0009] When the text type is printed text, text recognition is performed on the text area frame according to a standard text recognition model to obtain text characters;

[0010] When the text type is handwritten text, the stroke trend of the text in the text area box is determined, and the text characters are determined according to the stroke trend.

[0011] Preferably, the processed text image to be recognized is layered into a text layer and a background layer, including:

[0012] According to the grayscale values ​​of the pixels in the processed text image to be recognized, the region is divided to obtain a number of text image regions, and the grayscale value range of the pixels in each text image region is the same;

[0013] Detecting each text image region and determining the shape of each text image region;

[0014] A plurality of character radicals are preset, and the shape of the text image area is compared with the shape of each of the character radicals to determine the similarity between the text image area and each of the character radicals;

[0015] Determining whether the text image region is a character radical according to the similarity, and if the text image region is a character radical, determining the text image region to be a text layer;

[0016] If the text image region is not a character root, the text image region is determined to be a background layer.

[0017] Preferably, comparing the shape of the text image region with the shape of each character radical to determine the similarity between the text image region and each character radical includes:

[0018] Determine the center point of the text image area and the center point of the character radical;

[0019] The center point of the text image area is overlapped with the center point of the character radical, and the text image area is scaled and rotated to determine the coverage area ratio of the text image area and the character radical, and the coverage area ratio is set as the similarity between the text image area and the character radical.

[0020] Preferably, scaling and rotating the text image area to determine the coverage area ratio of the text image area and the character radical includes:

[0021] Determine the longest line segment passing through the center point of the character root and set it as the standard line segment;

[0022] Determine the longest line segment passing through the center point of the text image area, and set it as a comparison line segment;

[0023] Rotate the text image region with the center point as the rotation point until the comparison line segment and the standard line segment coincide with each other;

[0024] The text image area is scaled based on the standard line segment until the length of the comparison line segment is consistent with the length of the standard line segment, and the coverage area ratio of the text image area and the character radical at this time is determined.

[0025] Preferably, determining whether the text image region is a character radical according to the similarity includes:

[0026] A similarity threshold is set, and the similarity is compared with the similarity threshold;

[0027] If the similarity is less than the similarity threshold, determining that the text image area is not a character root;

[0028] If the similarity is greater than or equal to the similarity threshold, it is determined that the text image area is a character radical.

[0029] Preferably, the text layer is segmented to obtain a plurality of text area frames, including:

[0030] Binarizing the text layer to obtain a binarized text layer;

[0031] The binary text layer is segmented based on an image segmentation algorithm to obtain a number of text area frames, wherein the text area frames contain a number of adjacent characters.

[0032] Preferably, the text in the text area box is judged to determine the type of text in the text area box:

[0033] A plurality of character fonts are preset, and the text font of the text area box is detected. If the text font of the text area box is included in the character font, it is determined that the text type in the text area box is printed text;

[0034] If the text font of the text area box is not included in the character font, it is determined that the text type in the text area box is handwritten text.

[0035] Preferably, when text recognition is performed on the text area box according to a standard text recognition model to obtain text characters, the standard text recognition model is a convolutional recurrent neural network model.

[0036] Preferably, when the text type is handwritten text, determining the stroke trend of the text in the text area frame, and determining the text characters according to the stroke trend includes:

[0037] When the text type is handwritten text, extracting stroke features of the text in the text area frame, wherein the stroke features include stroke direction, stroke length, stroke curvature and stroke endpoint position;

[0038] Determining the stroke trend of the text according to the stroke features;

[0039] The boundary of each character in the text within the text area frame is determined, the text is divided into a number of character blocks, and the stroke trend of each character block is combined and analyzed to determine the text character corresponding to each character block.

[0040] The present invention also discloses an image text recognition device for an image scene, which is used to apply the above-mentioned image text recognition method for an image scene, and the device comprises:

[0041] An image acquisition module, configured to acquire an image of text to be recognized;

[0042] A text stratification module is configured to perform grayscale processing and filtering and denoising processing on the text image to be recognized, and to stratify the processed text image to be recognized into a text layer and a background layer;

[0043] A text type determination module is configured to segment the text layer to obtain a plurality of text area frames; and to judge the text in the text area frames to determine the text type in the text area frames, wherein the text type includes handwritten text and printed text;

[0044] The text recognition module is configured to perform text recognition on the text area box according to a standard text recognition model to obtain text characters when the text type is printed text; and to determine the stroke trend of the text in the text area box when the text type is handwritten text, and determine the text characters according to the stroke trend.

[0045] Compared with the prior art, the present invention has the beneficial effect that it proposes a method and device for image text recognition in image scenes, aiming to improve the accuracy and robustness of image text recognition, especially the processing capability of handwritten text and complex background images. By combining image preprocessing, text stratification, text type determination and targeted text recognition strategies, the present invention can achieve efficient and accurate image text recognition in complex scenes. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings in the following description are only embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying any creative work.

[0047] Figure 1 It is a flow chart of an image text recognition method for an image scene of the present invention;

[0048] Figure 2 It is a functional block diagram of an image text recognition device for image scenes of the present invention. DETAILED DESCRIPTION

[0049] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0050] like Figure 1 As shown, the present invention provides an image text recognition method for an image scene, comprising:

[0051] Get the text image to be recognized;

[0052] The text image to be recognized is grayed and filtered for denoising, and the processed text image to be recognized is layered into a text layer and a background layer;

[0053] Segment the text layer to obtain several text area boxes;

[0054] Judging the text in the text area box and determining the text type in the text area box, the text type includes handwritten text and printed text;

[0055] When the text type is printed text, the text area box is recognized according to the standard text recognition model to obtain text characters;

[0056] When the text type is handwritten text, the stroke trend of the text in the text area box is determined, and the text characters are determined according to the stroke trend.

[0057] In some embodiments of the present application, the processed text image to be recognized is layered and divided into a text layer and a background layer, including: performing region division according to the grayscale values ​​of pixels in the processed text image to be recognized to obtain a plurality of text image regions, and the grayscale value range of pixels in each text image region is the same; detecting each text image region to determine the shape of each text image region; presetting a plurality of character radicals, comparing the shape of the text image region with the shape of each character radical, and determining the similarity between the text image region and each character radical; determining whether the text image region is a character radical according to the similarity, and if the text image region is a character radical, determining the text image region to be a text layer; if the text image region is not a character radical, determining the text image region to be a background layer.

[0058] In this embodiment, the text image to be recognized is processed in layers, the text layer and the background layer are clearly distinguished, the interference of background information is reduced, and the accuracy of text recognition is improved. At the same time, the technical solution adopts steps such as gray value area division, shape detection, and similarity comparison to screen and recognize the text image layer by layer, making the entire processing flow clearer and more efficient. And by presetting character radicals and comparing them with the shape of the text image area, it can flexibly deal with text images of different fonts and sizes, enhancing the adaptability and versatility of the technical solution.

[0059] In some embodiments of the present application, the shape of the text image area is compared with the shape of each character radical to determine the similarity between the text image area and each character radical, including: determining the center point of the text image area and the center point of the character radical; overlapping the center point of the text image area with the center point of the character radical, scaling and rotating the text image area, determining the coverage area ratio of the text image area and the character radical, and setting the coverage area ratio as the similarity between the text image area and the character radical.

[0060] In this embodiment, by carefully comparing the shape of the text image area with the shape of the character radical, and matching by determining the coincidence of the center points and performing scaling and rotation. This method can capture the subtle features of the characters in the image, thereby avoiding the misrecognition that may be caused by ignoring these features in traditional methods. This technical solution is not limited by factors such as the direction, size, and slight deformation of the text image. Through operations such as scaling and rotation, it can adaptively match text images of different forms, thereby enhancing the robustness and adaptability of the algorithm. And by directly calculating the coverage area ratio of the text image area and the character radical as the similarity, the calculation process is simplified, and unnecessary calculations are reduced, thereby improving the overall processing efficiency.

[0061] In some embodiments of the present application, the text image area is scaled and rotated to determine the coverage area ratio of the text image area and the character radical, including: determining the longest line segment passing through the center point of the character radical, setting it as the standard line segment; determining the longest line segment passing through the center point of the text image area, setting it as the comparison line segment; rotating with the center point of the text image area as the rotation point until the comparison line segment coincides with the standard line segment; scaling the text image area with the standard line segment as the reference until the length of the comparison line segment is consistent with the length of the standard line segment, and determining the coverage area ratio of the text image area and the character radical at this time.

[0062] In this embodiment, the text image area is accurately scaled and rotated to ensure that the coverage area ratio of the text image area and the character radical reaches the optimal state, thereby improving the accuracy of character recognition. This technical solution can cope with text images of different angles and sizes, and dynamically adjust the position and size of the image area to perfectly match the character radical, thereby enhancing the adaptability and flexibility of the system. By setting standard line segments and comparison line segments, and using rotation and scaling operations, rapid adjustment and optimization of the text image area is achieved, thereby improving processing efficiency.

[0063] In some embodiments of the present application, determining whether a text image area is a character radical based on similarity includes: setting a similarity threshold and comparing the similarity with the similarity threshold; if the similarity is less than the similarity threshold, determining that the text image area is not a character radical; if the similarity is greater than or equal to the similarity threshold, determining that the text image area is a character radical.

[0064] In this embodiment, by setting a similarity threshold and comparing the similarity of the text image area with the similarity threshold, it is possible to more accurately determine whether the area is a character root. This method reduces the possibility of misjudgment and improves the accuracy of text image recognition. By directly comparing the similarity with the threshold, it is possible to quickly determine whether the text image area is a character root, avoiding complex image processing and feature extraction processes, thereby improving processing efficiency. In addition, this technical solution simplifies the complex text image recognition problem into a similarity comparison problem, making the operation process more concise and clear, and lowering the threshold for use.

[0065] In some embodiments of the present application, a text layer is segmented to obtain a number of text area boxes, including: binarizing the text layer to obtain a binary text layer; segmenting the binary text layer based on an image segmentation algorithm to obtain a number of text area boxes, wherein the text area boxes contain a number of adjacent characters.

[0066] In this embodiment, by binarizing the text layer and finely segmenting the binarized text layer based on an image segmentation algorithm, the text area can be identified more accurately, avoiding the problems of misidentification and missed identification that may be caused by traditional methods, thereby improving the accuracy of text extraction. The binarization processing and image segmentation algorithms in the technical solution are optimized for text characteristics, and can effectively shorten the processing time and improve the overall processing efficiency while ensuring the processing quality. This technical solution does not rely on documents or images in a specific format, can flexibly respond to text extraction needs in a variety of complex scenarios, and has strong adaptability and versatility.

[0067] In some embodiments of the present application, the text in the text area box is judged to determine the text type in the text area box: several character fonts are preset, and the text font of the text area box is detected. If the text font of the text area box is included in the character font, the text type in the text area box is determined to be printed text; if the text font of the text area box is not included in the character font, the text type in the text area box is determined to be handwritten text.

[0068] In this embodiment, by presetting character fonts and detecting the text fonts in the text area box, handwritten text and printed text can be accurately distinguished, avoiding misjudgments and missed judgments in traditional methods, and improving the accuracy of judgment. This technical solution does not rely on specific image processing algorithms or hardware. It only needs to preset character fonts to realize the judgment of text type, so that the system can adapt to different application scenarios and needs, and has strong adaptability and flexibility. Because this technical solution judges the text type by directly detecting the text font, compared with other complex image processing or machine learning algorithms, it has faster processing speed and higher efficiency, and can provide users with accurate text type judgment results more quickly.

[0069] In some embodiments of the present application, when text recognition is performed on a text area box according to a standard text recognition model to obtain text characters, the standard text recognition model is a convolutional recurrent neural network model.

[0070] In some embodiments of the present application, when the text type is handwritten text, the stroke trend of the text in the text area box is determined, and the text character is determined based on the stroke trend, including: when the text type is handwritten text, the stroke features of the text in the text area box are extracted, and the stroke features include stroke direction, stroke length, stroke curvature and stroke endpoint positions; the stroke trend of the text is determined based on the stroke features; the boundary of each character in the text in the text area box is determined, and the text is divided into a number of character blocks. The stroke trend of each character block is combined and analyzed to determine the text character corresponding to each character block.

[0071] In this embodiment, by extracting the stroke features of the handwritten text, such as the stroke direction, length, curvature and endpoint position, the uniqueness of the handwritten text can be captured more accurately, thereby more accurately determining the stroke trend and characters of the text, and improving the recognition accuracy. This technical solution not only considers the characteristics of a single stroke, but also determines the text characters by combining and analyzing the stroke trend of character blocks. This multi-level recognition strategy enables the system to maintain a high recognition capability when facing complex or ambiguous handwritten text.

[0072] like Figure 2As shown, the present invention also discloses an image text recognition device for an image scene, which is used to apply the above-mentioned image text recognition method for an image scene, comprising:

[0073] An image acquisition module, configured to acquire an image of text to be recognized;

[0074] The text stratification module is configured to perform grayscale processing and filtering and denoising processing on the text image to be recognized, and stratify the processed text image to be recognized into a text layer and a background layer;

[0075] The text type determination module is configured to segment the text layer to obtain a plurality of text area frames; and to judge the text in the text area frame to determine the text type in the text area frame, the text type including handwritten text and printed text;

[0076] The text recognition module is configured to perform text recognition on the text area box according to the standard text recognition model to obtain text characters when the text type is printed text; when the text type is handwritten text, determine the stroke trend of the text in the text area box and determine the text characters according to the stroke trend.

[0077] In this embodiment, by graying, filtering, denoising and layering the image, the text layer and the background layer are effectively distinguished, the background interference is reduced, and the accuracy of text recognition is improved. The text type determination module can automatically determine the text type (handwriting or printing) in the text area frame, and adopt corresponding recognition strategies for different types. This flexibility enables the device to be widely used in various image scenes and improve recognition efficiency. For handwritten text, the device identifies text characters by determining the stroke trend of the text in the text area frame. This method is closer to the characteristics of handwriting and helps to improve the recognition accuracy of handwritten text. In addition, the technical solution of the present invention can automatically process images and accurately identify text. Users do not need to perform complex preprocessing operations to quickly obtain text information in the image, thereby improving the user experience. It has certain modular characteristics, and each module can be optimized and upgraded independently, which provides convenience for subsequent function expansion and performance improvement.

[0078] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that they can still modify or replace the technical solution of the present invention with equivalents, and these modifications or equivalent replacements cannot cause the modified technical solution to deviate from the spirit and scope of the technical solution of the present invention.

[0079] The system provided in the above embodiment is only illustrated by the division of the above functional modules. In practical applications, the above functions can be assigned to different functional modules as needed, that is, the modules or steps in the embodiments of the present invention can be decomposed or combined. For example, the modules in the above embodiment can be combined into one module, or further divided into multiple sub-modules to complete all or part of the functions described above. The names of the modules and steps involved in the embodiments of the present invention are only for distinguishing the modules or steps, and are not regarded as improper limitations of the present invention.

[0080] Those skilled in the art should be able to appreciate that the modules and method steps of each example described in conjunction with the embodiments disclosed herein can be implemented with electronic hardware, computer software or a combination of the two, and the programs corresponding to the software modules and method steps can be placed in random access memory (RAM), memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, hard disks, removable disks, CD-ROMs or any other form of storage medium known in the technical field. In order to clearly illustrate the interchangeability of electronic hardware and software, the composition and steps of each example have been generally described in the above description according to the function. Whether these functions are performed in electronic hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered to exceed the scope of the present invention.

Claims

1. An image text recognition method for an image scene, characterized in that: The method comprises: Get the text image to be recognized; grayscale processing and filtering and denoising processing are performed on the text image to be recognized, and the processed text image to be recognized is layered into a text layer and a background layer; Segmenting the text layer to obtain a number of text area frames; Judging the text in the text area box to determine the text type in the text area box, the text type including handwritten text and printed text; When the text type is printed text, text recognition is performed on the text area frame according to a standard text recognition model to obtain text characters; When the text type is handwritten text, the stroke trend of the text in the text area box is determined, and the text characters are determined according to the stroke trend.

2. The image text recognition method for image scenes according to claim 1, characterized in that: The processed text image to be recognized is layered into a text layer and a background layer, including: According to the grayscale values ​​of the pixels in the processed text image to be recognized, the region is divided to obtain a number of text image regions, and the grayscale value range of the pixels in each text image region is the same; Detecting each text image region and determining the shape of each text image region; A plurality of character radicals are preset, and the shape of the text image area is compared with the shape of each of the character radicals to determine the similarity between the text image area and each of the character radicals; Determining whether the text image region is a character radical according to the similarity, and if the text image region is a character radical, determining the text image region to be a text layer; If the text image region is not a character root, the text image region is determined to be a background layer.

3. The image text recognition method for image scenes according to claim 2, characterized in that: Comparing the shape of the text image region with the shape of each character radical to determine the similarity between the text image region and each character radical includes: Determine the center point of the text image area and the center point of the character radical; The center point of the text image area is overlapped with the center point of the character radical, and the text image area is scaled and rotated to determine the coverage area ratio of the text image area and the character radical, and the coverage area ratio is set as the similarity between the text image area and the character radical.

4. The image text recognition method for image scenes according to claim 3, characterized in that: Scaling and rotating the text image area to determine the coverage area ratio of the text image area and the character radical includes: Determine the longest line segment passing through the center point of the character root and set it as the standard line segment; Determine the longest line segment passing through the center point of the text image area, and set it as a comparison line segment; Rotate the text image region with the center point as the rotation point until the comparison line segment and the standard line segment coincide with each other; The text image area is scaled based on the standard line segment until the length of the comparison line segment is consistent with the length of the standard line segment, and the coverage area ratio of the text image area and the character radical at this time is determined.

5. The image text recognition method for image scenes according to claim 4, characterized in that: Determining whether the text image area is a character root according to the similarity includes: A similarity threshold is set, and the similarity is compared with the similarity threshold; If the similarity is less than the similarity threshold, determining that the text image area is not a character root; If the similarity is greater than or equal to the similarity threshold, it is determined that the text image area is a character radical.

6. The image text recognition method for image scenes according to claim 1, characterized in that: The text layer is segmented to obtain several text area frames, including: Binarizing the text layer to obtain a binarized text layer; The binary text layer is segmented based on an image segmentation algorithm to obtain a number of text area boxes, each of which contains a number of adjacent characters.

7. The image text recognition method for image scenes according to claim 1, characterized in that: The text in the text area box is judged to determine the text type in the text area box: A plurality of character fonts are preset, and the text font of the text area box is detected. If the text font of the text area box is included in the character font, it is determined that the text type in the text area box is printed text; If the text font of the text area box is not included in the character font, it is determined that the text type in the text area box is handwritten text.

8. The image text recognition method for image scenes according to claim 1, characterized in that: When text recognition is performed on the text area box according to a standard text recognition model to obtain text characters, the standard text recognition model is a convolutional recurrent neural network model.

9. The image text recognition method for image scenes according to claim 1, characterized in that: When the text type is handwritten text, determining the stroke trend of the text in the text area frame, and determining the text characters according to the stroke trend, including: When the text type is handwritten text, extracting stroke features of the text in the text area frame, wherein the stroke features include stroke direction, stroke length, stroke curvature and stroke endpoint position; Determining the stroke trend of the text according to the stroke features; The boundary of each character in the text within the text area frame is determined, the text is divided into a number of character blocks, and the stroke trend of each character block is combined and analyzed to determine the text character corresponding to each character block.

10. An image text recognition device for an image scene, used for applying the image text recognition method for an image scene as claimed in any one of claims 1 to 9, characterized in that: The device comprises: An image acquisition module, configured to acquire an image of text to be recognized; A text stratification module is configured to perform grayscale processing and filtering and denoising processing on the text image to be recognized, and to stratify the processed text image to be recognized into a text layer and a background layer; A text type determination module is configured to segment the text layer to obtain a plurality of text area frames; and to judge the text in the text area frames to determine the text type in the text area frames, wherein the text type includes handwritten text and printed text; The text recognition module is configured to perform text recognition on the text area box according to a standard text recognition model to obtain text characters when the text type is printed text; and to determine the stroke trend of the text in the text area box when the text type is handwritten text, and determine the text characters according to the stroke trend.

Images