| Summary: | 碩士 === 國立中正大學 === 電機工程研究所 === 87 === The purpose of this thesis is to study the compression of multi-view images, by which any virtual intermediate views can be synthesized. Our coding scheme focused on not only the compression efficiency, but also providing sufficient information for rapid view generation that may follow. The coding scheme is block-based and in a similar principle as that of stereo MPEG.
There are two major parts in this thesis. In the first part, we develop an MPEG-like codec to code the multi-view images. Most blocks in the intermediate views are encoded with references to the left- and right-most views. The others, usually locating in the occlusion areas, are otherwise intra-coded. In this manner, not only low bit-rate can be achieved but also the occlusion information is retained.
Our codec is also featured of providing sufficient and reliable disparity information for the decoder. A sequential median filter is applied for the post-processing of disparities for more reliability. With them, the decoder is able to fast and randomly synthesize arbitrary views.
In the second part of this thesis, we interpolate the disparities from the received data and refer to appropriate images to synthesize arbitrary views. After patching the blanked area, the synthetic results look good. Our system has been implemented on a Pentium II 350 MHz PC. The processing time for each generation of intermediate view is about 0.04 sec (excluding the decoding time) if a image size is adopted. An Eye glasses system with infrared synchronization is used to view the stereo results. With appropriate hardware in computation, realtime display will be feasible.
Finally, the extension of our coding and synthesis architecture to multi-view video system is discussed as an end of this thesis.
|
|---|
