| Summary: | 碩士 === 國立清華大學 === 資訊工程學系 === 95 === In this thesis, we propose two efficient intra mode decision algorithms. One uses the image structure tensor, the local block gradient feature for image content. From the image structure tensor, we can determine the candidate directional modes that are most possible for representing the main edge direction in a block. After we find out the possible directional modes, we can just take these modes as our candidate modes for encoding the current block to improve the intra mode decision computation complexity. The second proposed algorithm is based on learning the mode conditional probability for the encoded intra mode for the current block given its simple gradient features and the encoded modes of its neighboring blocks. Based on this conditional probability, we can determine the most possible intra modes given the condition of the neighboring modes and the image gradient features of the current block. Similarly, we take these modes into our candidate modes for encoding the current block. In addition, we use the idea of intersection of candidate modes for combining these two proposed algorithms. Experimental results show our algorithms can efficiently reduce the computation complexity with negligible quality loss and with less bitrate increase compared to other previous intra mode decision algorithms.
|
|---|
