Key Technology And System Design Of2D-TO-3D Video Conversion

Three dimension movies can give people real sense of distance and bring people the perception of immersive. The movies shot by3D camera always need complex process, take long period and cost lots of money, which unable to meet the growing needs of people. The2D-to-3D video conversion technology using the relevant knowledge of computer graphics and image processing, can convert existing2D video to3D video automatically and quickly. So it cuts down the produce costs and can well solve the shortage of3D videos.This thesis purposes to build an effective2D-to-3D video conversion system. This thesis deeply researches the key technologies during the conversion, including depth map generation, virtual-view rendering, and image inpainting, then proposes relevant improved algorithms.This thesis realizes the technology of motion-based depth map generation. First calculate the motion vector by motion estimation. According to the principle of "the futher the smaller", convert the motion vector to relative depth and then create the original depth map. Then use morphological processing to remove block effect. Rectify contours in the original depth by referring to color-segment map. Expanding the depth map properly is proposed in the end to improve repaired image quality.This thesis improves a virtual-view rending technology. Considering the reality of2D-to-3D video, the positions of two eyes are fixed. So we can use relationship between depth and parallax to render the virtual-view image.This thesis improves an image inpainting technology based by Criminisi algorithm. The main idea of this algorithm is to fix the highest priority area first. In order to overcoming the disadvantage of Criminisi algorithm, this thesis proposes several improve methods as follows: Delay fixing the pixel which has a low confidence term. Convey data term using Sobel operators. Define the priority as weighted summation of confidence term and data term. Search the best matching block in left eye image. Use large block during searching while use small block during fixing. All these method can effectively improve the quality of repaired image.This thesis builds the2D-to-3D video conversion system using all these technology above. The system can convert the input2D video to red-blue or polarization format3D video. And accelerate it by using multi-thread technology. The conversion time rate of standard resolution video is under1:4.