Study On Binarization For Hybrid Document Image

With the advent of the era of information, the growing quality requirements ofmultiply for video and image urge us to improve the image saving and transmissiontechnology. Image compression, one of the main research direction of digital imageprocessing, research based on image coding methods to reduce redundancy. Theformat of the purpose is to achieve more efficient storage and data transmission.With the development of image compression theory and related standards, itdominated by lossless and lossless compression and loss compression of thecompression method for different image has obtained good effect. It has produced alot of mature and effective method for binary image compression like WBS coding,Huffman coding, LZW coding and so on. On this basis, we mainly study the premiseof binary image compression document binarization.Document binarization is a binarization process through the segmentation of grayimage based on the selection of threshold. It divided into two types: global thresholdand partial threshold according to the threshold selection method. The classic methodsof global threshold include between-cluster variance method(Otsu), minimum errormethod, etc. and the partial threshold include NIBLACK method, Bernsern method,etc. In allusion to some bad problems of Otsu method and Bernsern method, we madesome corresponding to their improvements. Otsu method is effect on distinctdocument and it satisfies our requirement for instantaneity. However, if there are somelocal information, such as some degradation of documents, this method can’t do itwell. Some person proposed to achieve the two-dimensional Otsu method and itsimproved algorithm by eliminating the repeat accumulative amount. In view of thetraditional two-dimensional Otsu method, we do the contrast of the improvedexperiment for the phenomenon such as real local information on account of the hightime complexity and comprehensive enough for considering local information. Theimprovement of Bernsern method is the use of area weighted threshold for different types of area. In this paper, we improve the inferiority of the algorithm throughexperiment. At last, we use the Laplacian sign figure method to deal with binarydocument images for some graphic and document mixed problem with bad effect, andwe improved it. After completing the segmentation of text and image which has thegray image with class of graphic and document mixed. We manage the text part withthe improved Laplacian sign figure method, and the image part with the improvederror diffusion method. It has obtained the good effect.