| Summary: | 碩士 === 國立臺北科技大學 === 電子工程系碩士班（碩士在職專班） === 104 === This paper proposes an algorithm applied to the lossless image compression by gradient prediction method. We want to reduce the data bulk by decreasing the image redundancy. However, data characteristics which were compressed affect compressing performance directly. In order to reduce the prediction loading, we proposed a simple gradient vote prediction algorithm. We construct the information by predicting three pixels around the current pixel, and classify texture trend of prediction unit. The majority decision and boundary detection are made with characteristics of neighbor pixels. Our gradient vote texture predictors can get higher coding efficiency in obviously changes boundary area in the image. On the other hand, binary mode encoding is a useful method to deal with the area which image pixels change slowly. In this thesis, we combine both advantages of gradient vote prediction and binary mode encoding. We use binary mode encoding when the image pixels of the area change slowly. Otherwise, we select regular mode to encode. Our gradient vote method is according to a proportional of the adjacent pixels statistics and strength to allocate the weight of the encoding pixels leads computational complexity is significantly reduced. However, complexly statist and estimate coefficient in error modeling has been offline trained and confirmed, no additional auxiliary information needs to be transmitted in coding prediction pixel. Our algorithm can significantly reduce the complexity and amount of information. It is able to obtain a well trade-off between computational complexity and prediction outcomes. The experiments depict that the proposed algorithm of the same type of comparison with the current forecasting techniques in prediction and encoders have shown the superiority of our proposed system.
|
|---|
