Research Of DIBR Hole Filling Method Based On Time Series

With the 3D movie "Afanda" set off a heat wave at home and abroad,3D technology has rapidly become the focus of the industry.The acquisition of 3D virtual viewpoint can be obtained by direct shooting method or some post-processing techniques.In the stereoscopic display,more viewpoints bring about more realistic viewing effects,which results in a higher cost by the direct shooting method,and the cost will grow exponentially with the shooting viewpoints' increasing.In recent years,Depth-Image Based on Rendering is a new synthetic technique of virtual view synthesis,which has the advantages of smaller bandwidth requirements,faster rendering speeds,etc.This paper presents the DIBR technique to generate a virtual viewpoint based on the original two viewpoints,and introduce the basic theory of stereoscopic display and the method of depth-image based on rendering.What's more,it also highlights the process of image in-painting after the rendering.Firstly,this paper introduce the fundamental principles and algorithms of DIBR and the hole problems existed in the target viewpoint.Then the front and rear frame is used to fill the target image holes.Lastly,the algorithm is improved by introducing a new method of hole filling based on time series.This new algorithm will be divided into two parts as the followings.The first one focuses on filling the hole based on the front and rear frame.That is,in order to deal with the problem of excessive hole regions,a new method of Gaussian smoothing has been put forward to reduce the hole number of virtual viewpoint drawing process.Then,the hole region of the target image is filled by the pixel information of the original 2D image.Compared with existing algorithms,this method has an excellent in-painting effect and the similarity with the original image can be as much as 75%.However,the hole filling effect on the complex texture might not seem so well.In the second part,we've proposed a new judgment method based on Structural Similarity Index Measurement(SSIM)to determine the successive frames of the 2D video frame,as the previous part based on frame around the image to fill complex hole area is limited.The range of the similar frames is settled by experiments,and the images of the original 2D ones are acquired by the time difference between the images as the reference and the hole filling link.With the help of the new point of view and combining with complementary methods proposed by the obtained reference image of the target image,we are able to repair the hole in the area.Experimental results show that the obtainment after filling effect of the image can be more real reduction of scenes;structure similarity to the original point of view can reach 84.46%,8.5 percentages higher than that of the traditional method.Experimental result shows that,compared with the traditional methods of image in-painting,the method filling holes basing on time series which this paper developed,has greatly improved the speed of hole-filling,and can restore the authentic background image to draw a new point of view which has high similarity.