The Hole Filling Inpainting Of 3-Dimensional Video Based On DIBR

With the development of multimedia technology and people's awareness of the world,3-dimensional video technology has gradually attracted people's attention.However,when3-dimensional image is synthesized by depth image based rendering(DIBR)technology,it is easy to generate large holes because of foreground occlusion.Therefore,how to fill the holes in the occluded area becomes a hot research topic of 3D video inpainting.In this thesis,according to the theory of the depth image based rendering technology,combined with the characteristics of the video sequences and the foreground/background segmentation theory,the 3-dimensional video image inpainting algorithm based on watershed algorithm and Gaussian mixture model is discussed.From the perspective of how to extract foreground/background effectively,the thesis discusses the3-dimensional video inpainting algorithm to improve the effect of three dimensional video inpainting.The main research result and contributions of this paper are summarized as follows:1.The thesis summarizes video image inpainting technology and the research status,analyzing the main problems and key technologies.At the same time,the principle of depth image based rendering technology is expounded in detail,with the image quality problems in the new view image.2.For the foreground and background segmentation in 3-dimensional video inpainting,the inaccuracy of the foreground object will easily affect the deficiency of the repair effect.The thesis combines the watershed algorithm with marker for image segmentation,making full use of the structural information of the depth image.In order to enhance the ability to distinguish the foreground objects,introducing K-means clustering to mark it in the gradient image.The experimental results show that the improved algorithm overcomes the defect of original watershed algorithm in image segmentation,which is likely to occur in the process of over segmentation,and it also can completely extract the texture information of the foreground object,make the repair results has better visual effect,and the peak signal-to-noise ratio(PSNR)is increased by 1 to 3 dB,compared to the original algorithm.3.The extraction of background images in the holes filling of 3-dimensional video images can easily cause the residual of foreground information and affect the repair effect.In this thesis,it uses the combination of simple linear iterative clustering(SLIC)and Gaussian mixture model(GMM)algorithm,and proposes a new background modeling method to realize the adaptive inpainting ofthree-dimensional video images.First of all,using SLIC to segment the image according to the color information,and make full use of the depth image to obtain a more complete foreground template.Then,according to the change of the foreground template and the depth value,the learning rate in the Gaussian mixture model is dynamically updated.Finally,according to the foreground template,the residual foreground in the background image is filled with Criminisi algorithm.The experimental results show that the proposed algorithm can extract the texture information of the foreground object completely and make the inpainting have a better visual effect.The peak signal-to-noise ratio(PSNR)improves by 1-5 dB compared with other algorithms.