| Summary: | 碩士 === 國立清華大學 === 電機工程學系 === 90 === Due to the advances in technology, the demand on transmission of multimedia becomes greater than ever. Many subscriber devices need the service of transmitting multimedia. According to the resolution of each device, the resolution of the video will need to adapt to the specific device. Hence a transcoder which will provide the service of translating resolution efficiently becomes more and more important. Recently, most researches concentrate on translating the resolution in the DCT-Domain. However, most papers just dealt with how to scale the picture by “2”. In this thesis we will propose a fast algorithm on resolution translation in DCT-domain for any scale factors.
We first propose a method to interpolate or decimate the DCT-domain sequence. This method won’t increase the original calculation capability, because the method needs to only multiply the original sequence by a scalar term, and this step can be combined into the quantization step in the transcoder. Then we take advantage of the symmetric property of DCT, preserve fewer low-frequency coefficients, and transform them back to spatial domain using smaller DCT. Then combine a proper number of them into a bigger block and transform them into DCT-domain using the proper DCT. Because we choose the size of block carefully, we will have many zero terms in the resulting matrices. Using this result we can save computation dramatically. Finally we compare our result with the bilinear scheme, and find that our scheme will have higher PSNR values and fewer calculation demand.
|
|---|
