Research Of Texture Synthesis Based Digital Image Impaiting Algorithm

As the development and progress in the field of network and digital technique, digital image is widely used in scanning, digital camera, digital TV and image transmission of mobile equipment. Image is stored, shared and transmitted in digital format, so a lot of factors will cause partial damage of image. In other application, such as removing object from the image will leave an empty hole, some techniques are needed to recover the full image. So the research of digital image inpainting algorithm is important and valuable.This paper introduces two kinds of image inpainting algorithms. One is analyzing image features based on PDE(partial differential equation). This paper does a deep research on both theory and practice of this algorithm. Another image inpainting algorithm is based on Texture Synthesis, This method is suitable to restore image with big damages and extend the application domain. But it is not perfect enough in a lot of facets. My work is to put forward improved algorithms based on previous research of AIEI and CIEI algorithm to improve the performance of the image restored.This paper introduces improved algorithms in three ways. Firstly, there are three improved ways of priority algorithm. The first one is based on p_laplace operator. This improved algorithm can remain the detailed information, without losing important information. The second improved algorithm is based on Euler's Elastica model, this improved algorithm introduce this model into decision of inpainting priority, and a factor p is used to adjust the inpainting along a better director between isophotes and grads. The improved algorithm combines the merits of both AIEI and CIEI, can restore the outline of the damaged object and remains the structure information. The third improved algorithm is based on curvature-driven diffusion (CDD), the inpainting order is not along the isophotes but also focus on the pixels which have big curvature, satisfies the connectivity principle. These three improved have common in the fact of connectivity, trying to remain the structure information of the damaged area.Then, two improved matching algorithm are introduced. The first one introduces the weight block, decide that different couples of pixels play different role based on their difference. And the second improved algorithm consider the confident data of every pixels , matching block is going to be found in the area originally known, and the couple of pixels with higher confident data should be matched as well as possible.At last, based on the theoretical analysis, lots of numeric experiments are performed, and the experimental results verify the improvements of the proposed algorithm.