Research Of Digital Image Inpainting Algorithms Based On Soft Computing

Reconstruction of missing or damaged portions of images, known as inpainting or retouching, is a very important topic in image processing and a hot spot in signal processing, computer vision and computer graphics. Its main work is to make a computer automatically inpaint appointed damaged region in a digital image according to the information of the regions around the damaged regions and characteristics of human vision.The conventional image inpainting method can be roughly divided into two kinds. One is partial differential equations (PED) based inpainting method, which mainly is applied to the recovery of damaged regions with small scale, such as blots, scratches and characters in an image. The other one is texture-based completion method, which obtains good effects when it is employed to damaged regions with large scale.Intelligent computing, also known as soft computing, was proposed by L. Zadeh in 1991. He classified artificial neural networks (ANN), fuzzy logic and genetic algorithm (GA) as soft computing areas, to distinguish from the traditional computing model. In recent years, with the rapid development of computer technology, more and more attention has been paid to soft computing. The research and application of inpainting the damaged region in small scale or large scale in a digital image will be discussed in this paper.This paper analyzes the traditional method of inpainting the damaged region and gives the theoretical principles and simulation results of the most representative model. After dicussing the traditional method, new inpainting methods are proposed and the contributions of this paper are as the following:1, Soft computing are employed in the digital image inpainting. An image inpainting method based on the self-organizing map (SOM) is first proposed.2, The above method is improved and an exemplar-based inpainting method using SOM is proposed. This method encourages linear structures to be synthesized to realize the connectivity principle of vision psychology.3, An image inpainting method based on GA is proposed. This method also encourages linear structures to be synthesized. The relatively large-scale damaged structural information can be repaired by this method according to human visual habits through GA optimizationThe merits and effectiveness of these three methods have been proven by the simulation results of this paper.