Threshold Image Segmentation Based On Edge Information

In practice, it is often required to segment uneven illumination image caused by uneven illuminance and uneven texture. Therefore, the industrial visual inspection system which based on high-precision close-up imaging must solve the problems in order to overcome the problems brought by uneven illumination. The paper does some research on it.First of all, on the basis of analysis of the existing algorithms' faults and the illuminance distribution characteristics of this kind of images, the thesis proposes a threshold image segmentation algorithm based on edge information and shows how to achieve it. At the same time, in accordance of the segmentation target, it proposes the technology solutions to multi-value segmentation based on edge information.Our algorithm takes the advantage of the image edge which is the image segmentation points.Secondly, the thesis proposes how to generate the threshold image based on edge information in detail. First, the image is filtered using sequential filtering, the edge image is achieved using eight-direction edge operator, and pseudo-edge is removed. Then the thresholds of image edge points are determined according to the high confidence level, at last the thresholds of non-edge points are determined with surface fitting using the edge points. In order to reduce the complexity in the surface fitting, the number of control points is reduced, and then Delauney triangulation is used to make division of regions, and finally the threshold image is rapidly fund through one-dimensional surface fitting for each region separately.Thirdly, the thesis researches on multi-level image segmentation. It shows an even illuminance color image multi-level segmentation method which integrates all the segmentation results of its gray image. Then we propose a multi-value segmentation algorithm based on Contour illuminance for the illuminance compensation. After that, in order to segment the uneven illuminance multi-level image, we propose the concept of the histogram of the edge image and the concept of marginal measure. Then the concept of the histogram of the edge image and the concept of marginal measure is proposed, classify the image edge with marginal measure, and generate each class edge's threshold image using the method of binary threshold image, and finally makes each pixel have the characteristic of multi-level image segmentation to achieve segmentation.Finally, the classical evaluation method is used to do some evaluation for our algorithm. A large number of experiment results show that this algorithm is effective for uneven illumination images and is stable for noise and fuzzy phenomenon.