Multiscale global sampling method for filling image void

Abstract

The invention provides a multiscale global sampling method for filling an image void, which mainly comprises the following steps: segmenting an input image into a geometrical layer and a texture layer, and recovering the geometrical layer of the image void area by tensor voting; carrying out multiscale grid segmentation on the image void area to form a multiscale image structure of the void area, defining a posterior probability model for the void filling problem on the multiscale image structure, and calculating the globally optimal solution of the posterior probability model by a Markov chain Monte Carlo method based on simulated annealing; and synthesizing missing image information of the image void area according to the void area integral sampling set corresponding to the globally optimal solution. The invention can search out a globally optimal image block to be attached to the void area, so as to fill or recover the void generated in the virtual scene object extraction and scene edition process, thereby solving the problem of unfavorable filling effect for a complex image or large void area condition due to the irreversible void filling process.

Description

technical field [0001] The invention relates to the technical fields of computer vision, image processing and virtual reality, in particular to a multi-scale global sampling method for image hole filling, which repairs the "hole" generated in the process of scene object extraction and scene editing, and ensures the image scene after filling Visually plausible. Background technique [0002] Image virtual scene generation is an important part of virtual reality and virtual scene generation. During the entire process of image virtual scene generation, the following situations may occur: after some scene objects are extracted from the image scene to form scene object materials, the original image scene will appear "hollow" phenomenon; Scene objects Before, it may be necessary to remove some unnecessary objects or objects from the scene; in the process of editing and blending scene objects, new hollow areas may be generated. In order to ensure the integrity of the image virtual...

AI Technical Summary

Problems solved by technology

[0009] According to the actual needs and key issues of image hole filling in scene generation, the purpose of the present invention is to propose a multi-scale global sampling method for image hole filling, find a global best image block and stick it to the hole area, fill or repair The "hole" generated in the scene object extraction and scene editing process overcomes the problem of poor filling effect on complex images or large hole areas caused by the irreversible hole filling process, ensures the integrity of the image virtual scene, and ensures the integrity of the image after filling. The visual plausibility of the image scene

Images