Image Damaged Regions Detection Based On Classification Model And Improved FCM Algorithm

With the development of digital image inpainting techniques, more and more researchers pay attention on it. In the existing image inpainting techniques, it is always need to localize damaged region firstly and repair damaged image. How to calibrate the repaired area of damage image is the premise for image inpainting. The damaged regions are even localized artificially when an actual damaged image is repaired. It is not suitable to repair mass damaged images, and a great deal of manual labor are wasted. The detection of damaged region is the key issue to achieve image inpainting and has been paid much attention.This thesis mainly targets the detection of the damaged area in digital image. The research is mainly for the scratch and handwriting damaged area detection, and other damaged areas detection in actual images. The damaged area detection is essentially a specific object detection and segmentation technique, it can regarded as the foreground segmentation, in which the damaged area as the prospects and non-damaged area as the background.Firstly, the threshold method and cluster analysis method of image segmentation are described. The objective evaluation index of image segmentation is given to evaluate the segmentation quality.Secondly, for the small damaged area such as scratches and writing existing in damaged image, a method is designed to detect and mark the scratches and writing damaged region by the classification strategy. This procedure uses the local characteristics of the damaged area caused by scratches and writing, and uses the histogram peak threshold detection method to detect the damaged area of sub-blocks. Then combining with the preliminary marked results of each sub-block area, the geometry and edge features of scratches and writing are used to complete the detection of damaged region at last. The experimental results show the correct detection rate of this method marked and scratch broken template, and gives the repair quality of the damaged image using different templates generated by this method and manually work. And the effectiveness and availability of this method is verified by the experimental results.Finally, for the larger damaged area in Tibetan mural image such as color loss situation, an image damaged regions detection method based on the improved FCM is designed. At the beginning, the validity of cluster is applied to adaptively obtained the initial numbers and centers of the clustering algorithm, getting a better number of clusters. In the meantime, the color feature vector and texture feature vector are used to segment the image. The cross-entropy distance measure is adopted to carry out FCM clustering. The damaged area is selected by human observing from the segment results, and used to detect and mark the damaged region by OTSU method. At last, given to the repaired image consisting of different damaged area, the repaired results of damaged image using the methods marked and manually labeled template are shown. And the availability of this algorithm is verified to detect the small damaged area such as scratches and damaged area of color fall off in the actual mural image by the experimental results.