A Document Image Binarization Method Based on Support Vector Machine

Abstract

The invention discloses a document image binarization method based on a support vector machine, including color image grayscale, document image block, improvement of local contrast of image block, feature parameter extraction, SVM threshold classification, image block splicing, stroke Eight steps such as width estimation and local binarization; the present invention uses the minimum mean method to grayscale the color image, and the resulting grayscale image has color independence; the defined local contrast can not only compensate for the influence of image brightness changes, The normalized contribution of all pixels in the image neighborhood to the local contrast of the image is also comprehensively considered; the SVM threshold classification method is used for high accuracy and reliability; the progressive scanning method is used for stroke width estimation, which has a relatively small impact on document image resolution changes. Good robustness; the invention can better preserve the stroke details of characters, and while effectively segmenting the foreground of characters, it can better suppress ink infiltration, page stains, textured backgrounds, and uneven illumination.

Description

technical field [0001] The invention belongs to the technical fields of digital image processing, pattern recognition and machine learning, and relates to a document image binarization method, in particular to a support vector machine (SVM)-based low-quality document image binarization method. Background technique [0002] Document Analysis and Recognition (DAR) technology has been widely used in printed characters and formula recognition, handwritten character recognition, document image segmentation, video subtitle extraction, text information retrieval and other fields, mainly including image acquisition, preprocessing, binarization, layout Analysis, character recognition, indexing, etc. Image binarization is one of the key processing steps, which directly affects the performance of the DAR system. However, binarization of such low-quality document images is extremely challenging due to factors such as image contrast, ink smearing, page stains, or uneven lighting. [00...

AI Technical Summary

Problems solved by technology

[0005] Researchers at home and abroad have also proposed many other methods, such as background estimation method, local contrast method, stroke edge detection method, gradient normalization and saliency map method, texture analysis method, Laplace energy method, error diffusion method, spectral Clustering methods and hybrid algorithms, etc., most of which have relatively high computational complexity, and cannot well solve the problem of binarization of low-quality document images affected by degradation factors such as ink infiltration, page stains, and background textures, or Can only be applied to some specific scenes (such as uneven lighting conditions)

Images