Image Denosing And Segmentation Based On Geodesic Distance

As an important preprocessing step in various digital image related applications,image denoising is still one of the most classical and challenging problems, So far,Many methods have been proposed for image denoising.The biggest advantage of interactive image segmentation，compared to traditionsegmentation methods，is that with the guidance of the user，problems whichpreviously were ambiguous in a semantic sense，now become clearly defined andsolvable．The major works can be summarized as follows:(1) An Shape-Adaptive window based on geodesic distance for non-local imagedenosing is proposed, which we calculate the geodesic distance in the similar windowover the first denosing result，then, we select a certain threshold to get the SA windowwhich is used for the mean pre-select. The effectiveness of proposed method is verifiedby the experiments of grayscale image compared with the method of patch pre-selectand SA-DCT pre-select.(2) A geodesic distance image segmentation method based on the patch similarity isdiscussed, which we use the MRF model on the Bayesian framework to calculates theweight for the geodesic distance. it is considers the patch information.The effectivenessof proposed method is verified by the experiments of nature image used the Gabor filterinformation and the single gaussian model.(3)We proposed a geodesic distance image segmentation based on the union model.We improve the original geodesic distance by introduce a factor. This factor couldeffectively amplify the difference between different groups and minify the differenceamong the group. And we use the average distance to get the result.The experimentsprove the effect of improved geodesic distance and the accuracy of segmentation getsfurther improved.This work was supported by the National Natural Science Foundation of China(No.61072106)，the Program for Cheung Kong Scholars and Innovative Research Team inUniversity(No. IRT0645) and the Fundamental Research Funds for the CentralUniversities(No. JY10000902032).