CN106503712A - One kind is based on stroke density feature character recognition method - Google Patents

Abstract

The present invention relates to image identification technical field, is based on stroke density feature character recognition method more particularly, to a kind of, including：Obtain images to be recognized；To obtaining Image semantic classification：Image slant correction is processed with correction chart picture and thresholding and obtains the single image of foreground information and background information；Analyzing and processing image：The textural characteristics in the ranks of analysis of the image, obtain the word matrix parameter of image；Segmentation figure picture：Image is cut based on the word matrix parameter, form several subimages, obtain the word block of image；Identification：Individual processing is carried out to word block, obtains the characteristics of image of word block, and described image feature is identified.The method word is simple, and discrimination is high.

Description

A Character Recognition Method Based on Stroke Density Feature

【Technical field】

The invention relates to the technical field of image recognition, in particular to a computer fault detection system and method, and a character recognition method based on stroke density features.

【Background technique】

With the widespread application of image acquisition equipment such as digital cameras, video cameras, and ultra-high-speed scanners, information in images has attracted more and more attention. The text embedded in the image is an important way of expressing the semantic content of the image, which can provide some important information that people need. For example, the text in an image can be an overview of the content of the image. If the text in the image can be automatically extracted and recognized, the computer can automatically understand the content of the image. Let the computer recognize the text in the image like a human being, which is of great significance for the storage, classification, understanding and retrieval of images and videos. It is mainly used in Chinese information processing, office automation, and its translation, artificial intelligence, etc. In the high-tech field, it has broad application prospects and commercial value. At present, the text in the image is generally only recognized through simple image segmentation processing, and cannot be adaptively adjusted according to the text features in the image, resulting in the low accuracy of the existing image text recognition method, which cannot meet the needs of practical applications.

【Content of invention】

In view of the above, it is necessary to provide a computer fault detection system and method, a text recognition method based on stroke density features, and the purpose is to solve the technical problem that the existing image text recognition methods have low recognition accuracy for text.

The object of the present invention is achieved through the following technical solutions:

A character recognition method based on stroke density features, comprising the following steps:

Obtain the image to be recognized;

Preprocessing of the acquired image: image tilt correction to correct the image and thresholding to obtain an image with single foreground information and background information;

Analyze and process images: analyze the interline texture features of the image, and obtain the text matrix parameters of the image;

Segmenting the image: cutting the image based on the text matrix parameters to form several sub-images, and obtaining text blocks of the image;

Recognition: process the text block separately, obtain the image feature of the text block, and identify the image feature; the image feature acquisition method is: calculate the text block frame, and add the framed text block p In the ×q lattice, project to the horizontal, vertical, 45-degree and 135-degree directions respectively, and each direction takes n values as features to form a 4n-dimensional feature vector.

Further, the preprocessing of the acquired image also includes performing image noise reduction on the recognized image to improve the accuracy of the recognition process.

Further, the image denoising process may adopt methods such as wavelet denoising method, morphological noise filter method, median filter method, adaptive Wiener filter method and mean filter method.

Further, the thresholding process includes a fixed thresholding method, an adaptive thresholding method, an Otsu method or an iterative method.

Further, the image in the image matrix is divided into the font in the matrix coordinates of the image is represented by the first pixel value, and the background is represented by the second pixel value, and the number of the second pixel value of each row in the matrix coordinates of the image is counted to obtain An array; count several line height parameters, take the average value of the parameters, and obtain the font size parameters.

Further, the recognition is based on a preset clustering algorithm to perform image segmentation processing on the segmented text sub-blocks, obtain the text information in the text block, and perform the recognition in the preset system text library according to the text information. Compare, determine the text in the image according to the comparison structure.

Further, the analyzing and processing the image also includes performing dilation processing on the text block.

Further, the identifying step includes performing normalization processing on the extracted text blocks before identifying them.

Beneficial effects of the present invention: the present invention analyzes the matrix row height texture feature of the image, calculates the matrix parameters of the image text, and then estimates the text font size parameter based on the text-related matrix parameters, and then quickly divides each text , and identify the text sub-blocks, which improves the accuracy of segmenting the text sub-blocks, thereby improving the accuracy of text recognition.

【detailed description】

A character recognition method based on stroke density features, characterized in that, comprising the following steps:

Obtain the image to be recognized; the image to be recognized can be any image that requires text recognition, and the image to be recognized can come from an external device. The image to be recognized may be an original image, or an image obtained by preprocessing the original image, and the image to be recognized may be in image formats such as jpg, bmp, png, etc.

Preprocessing of acquired images, including thresholding, thresholding, and skew correction. Thresholding: the thresholding includes a fixed thresholding method, an adaptive thresholding method, an Otsu method or an iterative method. The thresholding of the image is conducive to the further processing of the image, and the image with single foreground information and background information is obtained, which makes the image simple, and the amount of data is reduced, and the outline of the target of interest can be highlighted. Thresholding processing: Since the quality of the image to be recognized is limited by the input device, the environment, and the printing quality of the document, before the printed characters in the image are recognized, it is necessary to denoise the image to be recognized according to the characteristics of the noise to improve The accuracy of recognition processing, image noise reduction processing can use wavelet denoising method, morphological noise filter method, median filter method, adaptive Wiener filter method and mean filter method and other methods. Tilt correction: Since the scanning and shooting process involves manual operations, the image to be recognized input into the computer will have some tilt more or less. Before the printed characters in the image are recognized and processed, it is necessary to detect the image orientation and correct the image orientation .

Analyze and process the image, analyze the interline texture features of the image, and obtain the text matrix parameters of the image; divide the image in the image matrix into the first pixel value for the font in the image matrix coordinates, and the second pixel value for the background, and count Obtain an array of the number of second pixel values in each row in the matrix coordinates of the image; count several row height parameters, average the parameters and obtain font size parameters.

Segmenting the image: cutting the image based on the text matrix parameters to form several sub-images, and obtaining the text block of the image; before cutting the image, it also includes judging the arrangement direction of the text in the text area in the image, and the text text can be The block scans the pixels row by row and column by row to obtain the row spacing and column spacing of the text in the text block, and calculate the height variance of the text row and the width variance of the text column. The variance of the height of the text row is used to reflect the consistency of the height of the text row, and the variance of the width of the text column is used to reflect the consistency of the width of the text row. Then, it is judged whether the characters are arranged horizontally or vertically based on factors such as the consistency between the character spacing and the height of the character row or the width of the character column. For example, if the line spacing is greater than the column spacing, and the text row heights are the same, it is determined that the text in the text area is arranged horizontally. If the column spacing is greater than the row spacing, and the width of the text columns is the same, it is determined that the text in the text area is arranged vertically. Correct the segmentation results of text blocks, such as merging incorrectly segmented text rows or columns, or correcting the incorrect segmentation of the first letter and the second letter of the English language

Recognition: process the text block separately, obtain the image features of the text block, and identify the image features; use the text block after the layout analysis and single character segmentation operation to extract the text from the text area, It is also possible to expand the text block, and then use the text block to retain the text edge gradient and remove the interference of the local background gradient, thereby extracting each text from the text area and normalizing the extracted text The processing is to scale all the characters to a uniform size, and finally extract the features of each character for recognition. The image feature acquisition method is as follows: calculate the frame of the text block, and project it in the horizontal, vertical, 45 degree and 135 degree directions respectively in the framed text block p×q dot matrix, and take n values in each direction As features, a 4n-dimensional feature vector is formed.

The above is only a preferred embodiment of the present invention, and does not limit the present invention in any form. Although the present invention has been disclosed as above with preferred embodiments, it is not intended to limit the present invention. Anyone skilled in the art , without departing from the scope of the technical solution of the present invention, when the technical content disclosed above can be used to make some changes or be modified into equivalent embodiments with equivalent changes, but as long as it does not depart from the technical solution of the present invention, the technical content of the present invention In essence, any modifications, equivalent changes and modifications made to the above embodiments still belong to the scope of the technical solution of the present invention.

Claims (8)

1. a kind of based on stroke density feature character recognition method, it is characterised in that to comprise the following steps：

Obtain images to be recognized；

To obtaining Image semantic classification：With correction chart picture, thresholding process obtains foreground information and background letter to image slant correction The image of interest statement one；

Analyzing and processing image：The textural characteristics in the ranks of analysis of the image, obtain the word matrix parameter of image；

Segmentation figure picture：Image is cut based on the word matrix parameter, form several subimages, obtain the text of image Word block；

Identification：Individual processing is carried out to word block, obtains the characteristics of image of word block, and described image feature is known Not；Described image characteristic-acquisition method is：Word block frame is calculated, in the word block p × q dot matrix for adding frame, respectively To level, vertical, 45 degree and 135 degree of direction projections, each direction takes n value as feature, forms 4n dimensional feature vectors.

2. stroke density feature character recognition method is based on according to claim 1, it is characterised in that：Described to obtaining figure As pretreatment is also included to recognizing that image carries out image noise reduction to lift the degree of accuracy of identifying processing.

3. stroke density feature character recognition method is based on according to claim 2, it is characterised in that：At described image noise reduction Reason can adopt Wavelet-denoising Method, morphology scratch filter method, median filter method, adaptive wiener filter method and average Filter method is carried out.

4. stroke density feature character recognition method is based on according to claim 1, it is characterised in that：The thresholding is processed Including fixed threshold method, self-adaption thresholding method and Da-Jin algorithm or iterative method.

5. stroke density feature character recognition method is based on according to claim 1, it is characterised in that：To scheme in image array Font as being divided in the matrix coordinate by image represents that with the first pixel value background is represented with the second pixel value, statistical picture The number of every the second pixel value of row in matrix coordinate, obtains an array；To some row high parameters, parameter takes average primary system to statistics Meter, obtains font size parameter.

6. stroke density feature character recognition method is based on according to claim 1, it is characterised in that：The identification is based on pre- If clustering algorithm to cutting after word sub-block carry out image segmentation process, obtain the Word message in word block, and root Compare in preset system literal storehouse according to the Word message, according to the word compared in structure determination image.

7. stroke density feature character recognition method is based on according to claim 1, it is characterised in that：The analyzing and processing figure As also including carrying out expansion process to word block.

8. stroke density feature character recognition method is based on according to claim 1, it is characterised in that：The identification step bag Include after extracted word block is normalized and be identified again.