An Application Researching Of Edge And Shape Feature Information In Image Segmentation

In the actual image, due to the external factors, such as the difference of the camera angle or lighting range, etc., resulting in the situation of uneven illumination image or excessive exposure, etc. In some images, the target is small and the noise pollution is more serious which make the quality of the image decrease, and it has a great deal of influence for the results of segmentation processing. The traditional threshold segmentation method only use the gray information of image and it is less consideration for edge feature and shape feature and spatial related information, so these methods is poor for the image of target extraction which has no obvious boundaries between object and background gray distribution, even sometimes occurring the situation of serious fault points. In order to deal with the complex image better, some researchers have applied the method of multi-feature fusion to the segmentation algorithm, and obtained effective segmentation results.This paper aims at the image which is not uniform, small target and high demand kept of the edge information to segment. The edge and shape of the applied image are used as important feature information to improve the accuracy of segmentation. Considering the boundary region characteristics of image, the gradient information which can represent the edge characteristics is selected, and combined with the classical thresholding methods, a new threshold criterion function is constructed. In consideration of the shape contour feature, using the shape measure function which represents the shape characteristics, combined with the improved information entropy method, constructing the criterion function based on multi feature information.(1) To the classical exter-class, intra-class variance method (OTSU method), the OTSU method which is focused on the target is obtained through thought improvement of focusing on the objective. In the paper, the parameters in the algorithm is optimized using uniform measure function, and the algorithm optimized using weighted processing through constructing a gray gradient mapping function, the target is segmented using a new constructing criterion function with regional tectonic features of the image. At the same time, the accurately template computing gradient information is selected in the nine templates based on the principle of minimum variance, and the results is applied to the new criterion function to construct an accurate threshold segmentation criterion function with another gradient information.(2) For the classical maximum entropy threshold segmentation method, the gray level-gradient co-occurrence matrix is used for its improvement, and the improved two dimensional maximum entropy method is obtained based on gray level-gradient. Because of the new threshold segmentation algorithm proposed in this paper the gradient is only used to describe the edge information of the image, which has no effect on the threshold selection. In order to get more accurate image contour information, this paper, the improved maximum entropy method is weighted processing using the shape measure function based on the shape feature information of the image, and the new threshold criterion function is obtained.Experimental analysis and results show that combining edge and shape feature information to construct a new criterion function when using image basic gray information, and segmenting the image with illumination uneven, small target and high demand kept of the edge information, which can obtain the results of target image with internal information rich and clear edge contour, and also has good performance in anti-noise performance.