Stereoscopic Object-Oriented DT Mesh Coding

Nowadays, people are not satisfied with so little information from dull and ordinary video. Video images of three-dimensional will be the main development direction of multimedia technology. Human will be fully into the digital three-dimensional world in the near future because the stereo video and stereo image technology is developing with incredible speed. However, stereo video technology also encountered many problems when it is developing rapidly. One of the most main reasons is the stereo video and stereo image coding. It will put forward higher requirements of computer performance and network bandwidth when we process and transmit the stereo video, because the stereo video’s data is many times of ordinary video for displaying three-dimensional information. It will be the key to made the stereo video be practical and popular whether we can compress the stereo video efficiently. This thesis will adopt an effective coding scheme to compress the stereo image, the most basic component unit of stereo video. The experiments prove that the scheme of stereo image coding based on objects and DT mesh has a good performance to decode the stereo image with high quality when we need high compression ratio.Firstly, this thesis briefly describes the theory of stereo image and the coding principle. Then it puts forward the scheme of stereo images coding based on objects and DT mesh. The scheme of this thesis includes the following two aspects of content.(1) The first step is to extract the objects from the stereo image. Its purpose is to distinguish the objects between backgrounds. The thesis uses a scheme based on parallax figure. Due to different objects have different parallax’s information, we can extract higher quality objects through the use of threshold to extract the pixels belong to the objects, and optimize the objects by using the method of noise elimination, morphology processing, secondary matching, edge correction. Then though using the extracted object template, this thesis puts forward a simple method of extracting cover area. The theory of this method is narrow the scope of the cover area, then through the matching to find the cover area from the scope of the domain.(2) The second step is to use the scheme of stereo image coding based on objects and DT mesh to compress the images after extracting the objects. The objects and the background should be distinguished when extracting the feature points. The objects of the reference image use a method based on its edge and Grayscale distribution homogenization principle(MV principle) to extract the feature points. In this way, we can get more feature points, and the DT mesh structure is more matched with reference image. For the low attention of the background, we use the Harris method to extract feature points. For the purpose of eliminating the redundancy between the two images, we use all feature points to form the images’ DT mesh, then use the stereo matching and image prediction to let the reference image be close to target image. The experiments prove that the method of this thesis is better than JPEG, the image has less blocking artifacts. The experiments also prove that the decoding image which adopts the thesis’ method is better and more stereo than the image which adopts the method that just uses the initial points. And the efficiency of compression is higher than the method that applies MV principle to the whole image. The method for images with complex background is more efficient.