Research On Algorithm Of Segmenting Image And Text In Color Document Image

The Research on the image segmentation has begun since several dozen years ago. Thousands of algorithms of segmentation have been put out according to the many kinds of theory. Because there isn't any kind of general valid theory of segmentation, the algorithms available aim to solving specific problems and no algorithm of segmentation which is fit to all kinds of image has been achieved. Though efforts was paid to construct the model and segment the image using the model, no success was gained. Therefore , the custom algorithm of segmentation is the valid method.Document image becomes the main fundamental representation of information in the office documents and digital libraries etc. How to improve the compression ratio is becoming the hot focus of research, while the impression based on segmenting image is the kernel problem of compressing image of semantics. One of the valid means of improving the compression radio is to apply the different methods of compression according to the different area. A new algorithm of segmenting document image is proposed in this paper, which making use of methods of image processing and considering properties of document. The main works was as follows:First, it is discussed that the basic representation of image and the difference of different image formation. Then theory and aim of image process was discussed. Furthermore, some technologies, which can be used in the pre-process of segmenting document image, was discussed , for instance, smoothness, double color etc.Second, Through the discussion of transform relation between the color character and color space, the conclusion can be drawn that the proper selection of color space has important influence on the result of segmentation . At the same time some current general methods of segmentation are also discussed. They can not use the properties of document, and not segment the image properly . So a new segmentation model is put out which is fit to the divided-process compression strategy.Third, it is discussed that the specific realization of the algorithm of segmenting color document image based on multiscale. The destination of this algorithm is to divide the text area and picture area. Namely, leave out the text area and keep picture area. Firstly, through the pre-process(including transformation of color space, removing background noise and halftone), the texture of text area is enhanced ,while the texture of picture area is weakened. Then, to realize the goal of removing text area and improving the precision of segmentation, multiscale reduced image is applied. Then, to get the precise the marked image which is needed by segmentation, both the edge detection based on wavelet and the judgment of connection are used. Lastly, segmentation is finished by using the marked image and the copy of original image. Finally, the prospect of this algorithm was discussed . After segmentation , the next step is to compress document image highly. The experiment has prove that the compression based on image segmentation improves the compression ratio.