The Research On Multi-Information Fusion-based Virtual View Generation

In recent years, with the three-dimensional video technology in human life slowly spread, free viewpoint technology has been developing rapidly, rendering virtual view technology as one of the key free viewpoint asked to become today’s hot research topic. Virtual view rendering technology is to really restore the objects in three-dimensional space, and draw a virtual view image by image information and scene geometry. Virtual Viewpoint drawn mainly divided into two categories, one is for a simple scenario modeling virtual view rendering, and the other is an image-based rendering of complex scenes virtual viewpoint. The former has broad application in practical applications, but the latter because of the complexity of the algorithm, and the image information so as to obtain a single virtual viewpoint image quality is not very good, its application in practice, further research is needed. Therefore, in order to solve these problems, we proposed a novel and efficient image virtual view rendering algorithm based on multi-feature fusion in this paper:We proposed a multi-feature fusion-based image morphing algorithm. Extracting features and matching features in image morphing Algorithm are usually implemented through human-computer interaction, which leads to complicated implementation. The ghosting and blurring may occur in the occlusion area, which results in poorly experimental result in the same scene. In order to solve these problems, we proposed a novel image morphing algorithm based on multi-feature fusion in this paper. It extracts multiple types of image feature information to proceed with multi-feature fusion matching, which gets a properly distributed triangle mesh pair with correct correspondence. Then automatic image interpolation is achieved by conjunction of triangle mesh and image morphing. The experiment result shows that automatic extracted and matched features method effectively reduce the manual operation, and multi-feature fusion effectively restricts the ghosting produced at edges and occlusion area in image morphing. Through the questionnaire survey to the experimental results,91% of the participants think that the algorithm effectively improves the image morphing result.We proposed a kinect-based virtual view rendering algorithm, the depth image have voids and noise problems by Kinect somatosensory equipment acquired, It is difficult to directly use the virtual view rendering, the depth-image-based rendering needs camera calibration and this process is complicated and time-consuming. In order to solve these problems, we proposed depth image restoration by mean time domain filtering and joint bilateral filtering, Then we direct use kinect virtual viewpoint rendering combine special scene geometry information. Theoretic analysis and experimental results indicates that special scene geometry information effectively instead the camera calibration, and depth image restoration can use in a kinect-based virtual view rendering algorithm, and have a good effect of interpolation.