Repair Algorithm Based On The Decomposition Of Images And Video

Digital restoration is a technology using digital image and video processing to restore degradations existing in old photos and films such as scratch, blotch, flicker and so on. The basic principle of it is to use the information around broken image region or from the former and next video frame to fill in the inpainted region. It has mainly been used in the image transferring, image compressing, inpainting degraded images, restoring old films and so on. With the development of computer technology, how to inpaint images and videos automatically or half-automatically becomes an important task in image processing researches.In this thesis, we apply image decomposition technology to noised digital image and video inpainting, combining with the advantages of PDE based inpainting method and texture synthesis.On the aspect of image inpainting, an image decomposition algorithm named SID (Simplified Image Decomposition) is presented, which is based on total variation image restoration model and can reduce the computing complexity. The noisy gray image is decomposed into structure part ,texture part and noise part. Simplified total variation inpainting algorithms is used in structure part ,and pattern-based texture synthesis in noise part and texture part., finally the results are fused. Color images are separated into three channels according to lαβcolor space, which could avoid the relativity between channels. SID base image inpainting is used in each channel ,so the inpainting method of gray image can be applied for color image. The experimental results prove that this method overcomes the limitation of total variation inpainting algorithms and texture synthesis, and is robust in noise.On the aspect of video inpainting, decomposition based SROD blotch detection algorithm is designed, which can detect the broken regions automatically. Then spatio-temporal combination inpainting method was used on the detected region, which uses the spatio and temporal information. The experimental results prove that our blotch detection method is more robust in noise and lower fault-ratio than SDI and MRF, and our blotch inpaiting method can obtain better results than total variation and texture synthesis.