Research Based On The Document Image Layer Segmentation

Image segmentation by image processing into a crucial step in image analysis and a basic computer vision technology. This is because image segmentation, target separation, feature extraction and measurement parameters of the original image will be transformed into more abstract and more compact form, making more high-level analysis and understanding become possible. Therefore, image segmentation has been gaining attach great importance. In this paper, the mathematical description of image segmentation approach, give the image segmentation introduce commonly used in several ways, that is, gray threshold segmentation method, edge detection and tracking, as well as regional watershed algorithm based on the segmentation method.Thesis analyzes the characteristics of document images that document images are from some of the specific nature of a combination of regional block, usually including language block, line graph, continuous tone images and the halftone images . Among them, the characters and the line graph to preserve more image details and structural information with high spatial resolution characteristics of this part of the region do not ask for much of the color resolution; and continuous tone and halftone images are preserved more many of the color information, usually require a higher color resolution, spatial resolution but do not ask for much. This segmentation-based document image compression technology, the segmentation of different regions using different compression method can be a better compression effect. For document image segmentation layer effective thesis are the focus of the study.Optimize thesis presents a model of document image segmentation. The model will be split and layer block segmentation method combining thesis at the Document Image Segmentation optimization model, the main draw on the watershed algorithm L. Vincent morphological watershed-based segmentation methods and to improve the image of the layer to carry out partition. The basic idea of the algorithm are the image pixel gray-scale as the height of the image gray value of different regions on the peaks and valleys correspond to the basin, the edge of the basin is the so-called dividing line, dividing line of the origins of the name images are gradually Shen lake when a person: the lowest point of the image of advanced water, then water gradually diffuse over the entire valley basin, when the water level reached the edge of the basin would spill height. At this time the basin water spill at dams Agency to intercept, so until the entire image into the water, the dam was built in all the valley has become separated from the basin dividing line.At the watershed segmentation algorithm, the paper pointed out that the larger the watershed method, and the inadequacies of the largest segmentation is often because of noise and other anomalies such as the gradient of the local impact of over-segmentation. Application of over-segmentation algorithm is enough to make the results useless. Therefore, in practical applications, using an improved watershed algorithm, that is, first of all original images on smooth, will smooth the image using the watershed after the transformation, the same labeling of the pixels belong to the same water basin, and will be different from the storage from the same water basin, the pixel marked as a watershed point, and this has been the initial image segmentation results; Finally the application of gray-scale homogeneity criteria and guidelines for the border regions of intensity integration, in order to address the problem of over-segmentation. The experimental results show that the method can be a precise and meaningful segmentation results.The last article introduced the document image layer segmentation segmentation optimization model experiments, and give an example of image segmentation situation. On the results of segmentation with a single layer of block segmentation and segmentation results are compared. So that optimal partition layer partition model has a better effect.