Multi-Scale Decompositions With Local Weighted Prior

With the popularity of all kinds of digital products in our daily life, digital image processing has received widespread attention in recent years. As the fundamental problem in image processing, multi-scale image decomposition has been extensively studied.Many recent computational photography techniques decompose an image into a smooth base layer containing large-scale variations and a residual detail layer capturing the smaller scale details in the image. In many applications, it is important to control the spatial scale of the extracted details; in other words, it can be desirable to manipulate details at multiple scales.Aiming at this problem. Firstly, we have extensively studied the traditional edge-preserving algorithms such as bilateral filter, guided filter, weighted least squares etc. Based on local area weighted prior method was proposed to keep the image edge. Secondly, the paper develops a new local weighted prior that is able to describe some structures of an image, e.g., smooth regions, tinny details, etc. This adaptive weighted prior with TV and LO regularization is able to preserve large-scale structures and remove tiny details. Finally, the proposed method is solved by alternating iterative minimization, which has fast convergence property.Experiments show that our method based on the local weighted edge-preserving smoothing algorithm proposed in this paper has achieved better results than the traditional edge-preserving smoothing algorithm, which can keep the image edges very well and avoid visual artifacts in image processing at the multi-scale decompositions and isolated noise. Moreover, our image decomposition algorithm is faster than the traditional edge-preserving algorithms. It is an efficient algorithm which is able to extract details at the arbitrary multi-scale.