2D-3D Conversion Technology Guided By 3D Quality

3D industry is developing rapidly. Academia and general public have more and more interest on the 3D display. However, because of the complexity of 3D videos production, compared with 2D videos, the lack of 3D content has become a bottleneck that 3D industry is popularized. In addition, people’s interest in 3D displaying grows. Many people want to convert their own 2D videos into 3D videos with the help of some tools for the better viewing experience.As the fact that vast 2D videos exist at present, the method that 2D videos are converted to 3D videos by 2D-3D video conversion technology can ease the problem about the lack of 3D videos. According to the degree of manual intervention,2D-3D video conversion can be divided into three classes:manual conversion, semi-automatic conversion and automatic conversion. Manual conversion can get the highest video quality, but it expends expensive time and human costs, leading to the difficulty of popularization. Semi-automatic conversion just needs small amount of manual intervention, where automatic conversion has no use of manual intervention. The two kinds of conversion technology also can get high video quality with the appropriate algorithm. Therefore, they have gained wide attention from researchers.This paper proposes the method of key-frame selection and strategy of depth propagation based on the image distance. Firstly, we discuss the definition of image distance. This paper adopts different distance in different process for the best result. Secondly, we choose two kinds of global features, HSV color histogram and GIST feature, as the image distance of K-means clustering. And the results are evaluated by MSE. Thirdly, this paper proposed two kinds of depth propagation strategies. The first one is that depth propagates from key-frame to all non-key-frames. It saves the time and the calculated amount, and still gets a good quality of depth images. The second one is that all frames are sorted strictly in a cluster based on Prim’s algorithm. It guarantees that the total distance of depth propagation is minimum and then the error of depth propagation is minimum. This method gets more accurate depth propagation and higher quality of depth images. With a series of experiments, the two strategies can get less error in depth propagation compared to other methods, which proves our methods’ effectiveness.This paper applies the idea of proposed method in 2D-3D video conversion to 3D reconstruction. The idea of clustering is used in images selection and points cloud denoising. It improves the efficiency and quality of 3D reconstruction.This paper discusses quality evaluation criterion of 3D videos and adopts the suitable evaluation methods to test the proposed method. The result proves the strength of our method. It states that our methods can improve the efficiency of 2D-3D conversion and get a better quality of 3D videos.