The Research Of Image Upsampling Algorithms

In recent years, display terminals are diverse and their resolutions are becoming higher and higher. The efficiency and quality of image super-resolution algorithms are in demand. Among various image super-resolution algorithms, the approaches of linear combinations of kernel and samples are efficient and easy to be parallelized. However they tend to introduce blurring or block artifacts, and their computational costs will increase rapidly when the degrees of polynomials in-crease. While the adaptive methods based on feature analysis can produce better effects and arouse more attention.In this paper, several image super-resolution algorithms are proposed to address above prob-lems.1a new unbiased quadratic interpolation kernel based on B-spline is proposed. After careful theoretic analysis and experiments, we draw that the proposed algorithm can resize the images with comparable quality with Catmull-Rom algorithm. While it has improvement on computing efficiency and less undershooting/overshooting effects.2, a new adaptive bilateral filtering algo-rithm for image denoising is proposed. It adaptively adjusts the parameters in the filtering kernel function according to the gradient of edge in the image. As a result, it can remove the noise while preserve the feature edges effectively.3, an adaptive image super-resolution algorithm based on edge extraction is proposed. The feature edges are extracted in the sub-pixel level. The interpola-tion takes both the feature edges and edge blending areas into account. The experimental results show that the algorithm can generate satisfied results.