| Summary: | 碩士 === 南台科技大學 === 資訊工程系 === 97 === Motion estimation (ME) can remove the information of temporal redundancy to reduce the bit rate for video coding. Due to the great computation required in motion vector estimation, many simplified search algorithms have been proposed. Among them, the one-bit transform (1BT) scheme achieves the best efficiency with the lowest computational complexity. Thesis proposed a dynamic threshold to form the binary images with respective to each macro-block images. Motion vectors are calculated through a series of Boolean operations, such as “XOR”, “AND”, and/or “OR”. In addition, a multi-bit transform (MBT) method is also proposed to perform the weighted non-matching points measure (WNNMP) using a number of binary images. Moreover, for subjectively compare the performance of ME methods, this thesis proposed a measure called the normalized degradation of PSNR (NDPSNR). The efficiency of the motion estimation algorithms would be evaluated by NDPSNR measure. The experimental results show that the proposed approaches can achieve the better performance.
|
|---|
