Self adaptive two-valued method, device for file

Abstract

The present application relates to a method, device and storage medium for adaptive binarization of documents. The method for performing self-adaptive binarization on a gray-scale document image of the present invention includes: a division step, dividing the gray-scale document into blocks; a first determination step, determining the background block and the background block among the divided blocks according to the characteristics of the blocks Text block; the second determining step is to determine the background pixels in the pixels included in each text block determined in the first determining step; the first calculation step is to calculate a block threshold surface representing the threshold of each block, wherein, based on the first All pixels included in the background block determined in a determining step calculate the threshold value of the background block, and the threshold value of the text block is calculated based on the background pixels included in the text block determined in the first determining step, the background pixels are in the second determined in the determining step; and a binarization step of binarizing the grayscale document image using the block threshold surface calculated in the first calculation step.

Description

technical field [0001] The present invention relates generally to document image processing, and more particularly to optical character recognition (OCR). More specifically, it relates to a document adaptive binarization method, device and storage medium. Background technique [0002] Binarization of document images is the first step in document image analysis systems such as optical character recognition systems. The output of the thresholding operation is a binary image, where one state represents the foreground object, ie printed text, and its complementary state corresponds to the background. Binarization methods can be divided into two categories: global and local thresholding segmentation techniques, see .D. Trier and A.K. Jain. Goal-directed evaluation of binarizaton methods, IEEE Transactions on Pattern Analysis and Machine Intelligence, 17(12): 1191-1201, 1995, Yibing Yang and Hong Yan. Anadaptive logical method for binarization of degraded document images, Patt...

AI Technical Summary

Problems solved by technology

[0005] 1. Changes in background intensity due to uneven illumination and improper storage;

[0006] 2. Very low local contrast due to blemishes, stains and shadows in the capture process of document images;

[0007] 3. Non-stationary and signal-related noise;

[0008] 4. There is a change in the contrast between the foreground and background areas of the image;

[0010] 6. Grayscale images have low resolution

[0020] 1. The global threshold segmentation technique is suitable for high-quality document images, but it does not work very well for documents with poor quality;

[0021] 2. The local threshold segmentation technique does not work very well for small-sized images, because the image must first be sampled and reduced into smaller blocks, and then expanded

Boundary padding can seriously affect the final result;

[0022] 3. Some local average threshold segmentation techniques such as the Niblaek method often amplify noise and tend to misclassify large background areas as text;

[0023] 4. These techniques often require the use of edge detection techniques, thinning methods and / or post-processing to remove "ghost" objects;

[0024] 5. The local threshold segmentation technology requires multiple image readings, which is very slow and not suitable for high-quality images

Segmentation methods can outperform it in terms of sophistication if it becomes too complex for improved

Images