Image restoration based on Partial Least Squares regression and Wavelet Bi-cubic ratio interpolation

Abstract

This paper presents a novel super-resolution image restoration algorithm based on Partial Least Squares regression and the Wavelet Bi-cubic ratio interpolation. By wavelet transform, the original low-resolution image can be decomposed into high-frequency and low-frequency sub-images. Before reconstructing images by inverse wavelet transformation, high-frequency sub-images can be interpolated according to the bi-cubic interpolation algorithm. At the same time, the high-resolution image can be obtained from the original low-resolution image by bi-cubic interpolation. Firstly, fuse the high-resolution image with reconstructed image by inverse wavelet transformation. Then, take the fused image as training samples to restore original images by partial least squares method. At last, we compare these results, which are obtained from the algorithms based on the interpolation (such as the nearest neighbor interpolation, the bilinear interpolation, the bi-cubic interpolation and so on). The experimental data shows that the super-resolution restoration algorithm based on partial least squares and wavelet bi-cubic ratio interpolation algorithm can get better effect than the traditional algorithms.