Exempler Based Methods For Linear Target Extraction And Post-processing

In digital image processing, specific target identification and detection is the key to automatic target recognition technologies. Linear target as one part of the features of the digital images determines the quality of target recognition in a large extent. Digital image during the linear feature extraction due to the methods their own relationship, often there are some flaws.In this paper, take the retinal vessels and roads as two kinds of linear targets, focused on the example based methods for linear target extraction and post-processing. The main idea is to through to comparing the target has been marked, which can get the statistical model of the target.The main content of this paper is as follows:1)Proposed a post-processing method on the results of road extraction. First, delete the fake road segments according to length, curvature and other characteristics. Then, we connect road segments by calculate the shortest path. After the steps mentioned above, we can get a full result.2)Improve the problem that small retina vessel segmentation of existing methods is not satisfactory. Proposed the ratio as a feature of retina vessels. Detect small vessels on the existing segmentation results and modeling. Finally, we get the new segmentation result by continue to add matching points in the neighborhood of the sample. And we also made an experiment to compare our results with the original segmentation results.3)A post-processing method of retinal vessel segmentation was proposed with the non-local means idea. Firstly, we got the sample of vessels by calculate the ratio which was mentioned in1).Secondly, we got the sample of background by detecting the smooth region. Then, the weights of every pixel was calculate using the non-local means so that we got pixels’ probabilities of vessels and background. Finally, post-processing is done by comparing each pixel’s probability of vessels and the probability of background.This work was supported by the National Natural Science Foundation of China (No.61072106), the Program for Cheung Kong Scholars and Innovative Research Team in University(No.IRT0645) and the Fundamental Research Funds for the Central Universities(No.JY10000902032).