MRF sample block image restoration method by use of edge statistical characteristics

Abstract

The invention discloses an MRF sample block image restoration method by use of edge statistical characteristics. The method comprises following steps of determining a damage region to be omega for a to-be-restored image I; dividing the to-be-restored image I into an edge characteristic image Iedge and a non-edge characteristic image Inon-edge; counting offset mapping between similar sample blocks, and using the offset mapping as candidate tags for following optimization solving; by use of Curvelet conversion, extracting eight-direction characteristics of the to-be-restored image I; constructing a global energy equation keeping neighborhood continuous consistency; by use of the selected candidate tags and a multi-tag graph-cut algorithm, solving a global energy extreme value; and copying information of a tag corresponding to each node into an unknown region so as to obtain a restored image IR. According to the invention, continuity of structure parts in the restored image can be effectively kept; the restored image looks quite natural and quite meets vision requirements of eyes; and the method is especially suitable for restoration of real pictures or synthesized images with complex textures and structure characteristics.

Description

technical field [0001] The invention relates to an image restoration method based on sample blocks, in particular to an MRF-based image restoration method. Background technique [0002] Digital image restoration is a technology to repair the lost information according to certain rules according to the known information in the image to be repaired. Its main purpose is to make the restored image look coherent and natural. With the development of digital image processing technology, digital image restoration technology has become a research hotspot in computer graphics and computer vision. It has important applications in the protection of ancient cultural relics, production of special effects for film and television, image lossy compression, and removal of specific objects. value. At present, digital image inpainting techniques are mainly divided into three categories: diffusion-based methods, sparse-based methods, and sample block-based methods. However, sample block-based ...

AI Technical Summary

Problems solved by technology

However, these methods cannot maintain the continuity of the repaired image structure well for damaged images with less structural information.

Document 2 only uses the known information in the adjacent position to fill in the missing information, and introduces the gradient information to construct the global energy equation, but because the selection of candidate labels is not suitable enough, it still cannot repair the damaged image well

Documents 3 and 4 count the offset mapping between similar sample blocks, and select candidate labels accordingly. However, for damaged images with less structural information in known regions, the continuity of the repaired image structure is still not well maintained.

Images