Depth Map Generation Based On Depth Cues

Compared with the traditional 2D film, more and more people like the 3D film because 3D film can bring stronger visual impact and visual enjoyment. However, making 3D film is much more complex than 2D film as well as it costs much in money and production period. Existing 3D film is far from meeting the needs of the huge market demand, so it is important to make the existing 2D film conversion to 3D film which can make the most of the abundant 2D film resources. In addition, people also want to watch the corresponding 3D version to recall their memory. In the process of 2D to 3D conversion, the generation of the depth map is the key step, which directly affects the later 3D effect. Therefore, it is great important to investigate how to generate the high quality depth map in 2D to 3D conversion.In this paper, the method of depth map generation based on depth cues in 2D to 3D conversion process is discussed. Detailed introduction of a variety of depth cues and depth map generation methods based on each depth cue, in order to lay the theoretical foundation for the follow-up research work. The main research work of this paper is as follows:1. A depth map generation method based on block DCT is introduced. The method firstly divides the input image into blocks and performs DCT transform in each block. Then a depth map is generated according to the obtained coefficient. Lastly, joint bilateral filtering is applied in the depth map. On the basis of this method, a new method of generating depth map based on sliding window DCT is proposed, and the new method solves the problem of the block effect of initial depth map generated by the former method. When the depth map is filtered in the late stage, in the similar visual situation, the window size of the joint bilateral filter is significantly smaller than the former, and the computation time is also saved.2. A new method for generating depth map based on entropy is proposed. The method used the entropy generates the focus region of the image as the foreground object. A single depth value is used to assign the foreground object values and the depth gradient hypothesis model is used to assign the background object. Lastly combing the foreground object and the background object to generate the depth map.3. Depth map generation using hybrid depth cues is discussed in the last part. Combing the saliency map obtained by visual saliency cue and the initial depth map generated by the sliding window DCT method to extract the foreground object, which assigns the depth distribution of foreground and background and generate the depth map. Considering the effects of atmospheric scattering depth cue, and combined with the entropy-based method, these two depth maps are generated respectively. Through the integration of different weights, improving the shortcoming of lack of depth information was extracted by a single depth cue, which improves the quality of the depth map.