Variable-block image repair method based on saliency map

Abstract

A variable-block image repair method based on a saliency map comprises the steps that (1) a selective attention model simulating a living body visual attention mechanism is used for generating the saliency map of an image to be repaired; (2) items of credibility, structural information and saliency in blocks are calculated and the repair priorities are obtained, wherein each point on the boundary of an unknown region serves as the center of the corresponding block; the block of which the point with the highest priority serves as the center is used as the block to be repaired; (3) according to pixel mean value changes, gradient changes and saliency changes around the block to be repaired, the size of the block is adjusted; (4) a known region is searched for the most matched block to fill pixels in the block to be repaired; credibility and saliency of the repaired pixels are updated; the step (2) to the step (4) are cycled until the image is repaired. According to the method, saliency information of the image to be repaired is used for improving the priority of repair, so that the problem of avarice of traditional repair is greatly solved; meanwhile, the size of the block can be adjusted according to the surroundings of a region to be repaired and different repair requirements are met.

Description

technical field [0001] The invention belongs to the fields of computer vision and computer graphics, and in particular relates to a variable block image restoration method based on a saliency map. technical background [0002] Image inpainting technology has a wide range of applications, mainly including damaged photo restoration and digital visual effects. Due to various factors such as storage conditions or man-made, some works of art or old photos with extremely high collection value and historical value have cracks and scratches. Therefore, it is of great significance to restore these cultural relics and data. Through digital image repair technology, only need to determine the area to be repaired, the computer can repair it according to the specified algorithm, avoiding a lot of time spent on manual repair. At the same time, applications in the field of digital visual effects production and virtual reality mainly involve object removal and scene editing. Specific appli...

AI Technical Summary

Problems solved by technology

However, these methods are limited by two points: the greediness of their repair and the influence of block size selection on the repair effect

Although there have been some methods based on global optimality trying to solve its greediness, the results show that these methods only alleviate its greediness, and at the same time bring a lot of computational complexity

Similarly, on the issue of block size selection, if the block selection is too small, the local texture characteristics cannot be fully considered. If the block selection is too large, it may cause the block to contain too much structural information, which is not conducive to restoration.

At the same time, blocks that are too large are also susceptible to parallax and small-angle rotation

The general method is to artificially select the block size that can contain the largest resolvable texture in the known area, and in many cases the same block size cannot meet the requirements of all unknown area repairs.

Images