Kernel Regression Based On The Structure Tensor Is Non-uniform Interpolation Algorithm And Its Applications In Image Processing

With the development of information technology and more applications of image processing, interpolation, as a means of image processing, has played a very importan: role. Many problems can be solved by interpolation, such as image zooming, image denoising and image super-reconstruction. The traditional interpolation methods are effective in dealing with data on regular nodes, but in practical, the sample points are often irregular, because of the error caused by various aspects, such as motion blur, data loss and other reasons, therefore irregular interpolation has very practical significance.This paper first provides a detailed introduction to irregular interpolation based on kernel regression, and then their defects are analyzed. A new kernel regression based on structure tensor for irregular interpolation is proposed. At the same time, the paper applies the new algorithm to image zooming, image denoising and image super- reconstruction.The major results of the research results are as follows:1. Structure tensor based kernel regression for irregular interpolation is proposed. Firstly, the latest kernel regression for irregular interpolation is introduced, and then comes the analysis of the shortcomings of the bilateral kernel regression and steering kernel regression. And then structure tensor based kernel regression for irregular interpolation is proposed. In the new algorithm, the structure tensor is used to design adaptive kernel function which adapts better in the image edges, because structure tensor extracts image structure information. It makes the new irregular interpolation more effective. Experimental results show that the new algorithm has its superiority.2. The new structure tensor based kernel regression for irregular interpolation algorithm is applied to image zooming, image denoising and image super-resolution reconstruction .First, image zooming, image denoising and image super-resolution reconstruction are described and transformed into irregular image interpolation problems, and then the corresponding interpolation models are built. Then, the new algorithm is used. The experimental results show that the new algorithm has better performance in estimating the structure and texture of the images.