Edge reinforcment method by using segmentation

The success of image processing, computer vision, pattern recognition applications mostly depends on the detection of object edges. Although there is a lot of research on this subject, there are still no strong algorithms for edge detection, because factors such as reflection, noise, threshold value affect the accuracy. In this study, as a result of examining these algorithms in detail where they have failed at the trailing edge, a new method, edge reinforcement by using edge segmentation (SKG), is proposed for detecting and enhancing the edges. By this method, detecting object edges process discussed from a different viewpoint. In the method, instead of detecting the edges separating the regions by viewing sudden changes in the pixel densities, applied by the other edge detection algorithms, an answer was sought to determine whether the edges can obtained from the regions. In this study developed by using MATLAB software, two reference regions (RB) which have all the information necessary for exploring the missing tips of the edges are formed by using the sides of the edges that first obtained by known methods like Sobel, Prewitt, Roberts, Canny. Then, the estimated directions of the edge to be continued are determined by considering the direction of the reference edge and new regions are formed in the orientation of these directions. In the next step, with statistical methods, by comparing created regions and the RB, the direction and the region with the greatest edge possibility are determined. In the last step, by using similar to RB pixels of the obtained area, curved edge of the region were drawn. The presented edge reinforcement algorithm was applied to different images with different contrast values and the angiography data set from the Berkeley Segmentation Dataset and the Hypermedia Image Processing Reference (HIPR) database and compared with the results of popular edge detection methods. The results show that the developed algorithm can detect the edges that other methods have found and complete the missing edges.