无损压缩中基于块的梯度预测模式

Gradient-adjusted Prediction Mode of Lossless Compression Based on Block

为应对自然图像纹理属性的多方向性要求，提出一种基于块的梯度预测模式。对不同的块采用水平、垂直、左斜以及右斜4种扫描方式，并根据图像纹理属性自适应选择扫描方式。块中每个像素则通过加权不同扫描方式下的相邻像素预测得到。实验结果表明，与CALIC单一水平扫描方式的梯度预测方法相比，该方法能有效提高预测精度，且预测残差的零阶熵可降低3％～12％。

For multi-directional texture properties of natural images, this paper proposes a new Gradient-adjusted Prediction （GAP） method of lossless compression based on block. For each block, four scan modes are adopted including horizontal vertical, right down and left down direction and the scan mode is selected by an adaptive coding scheme based on the texture property of the image. Each pixel within a block is predicted by a weighted sum of its different neighbor pixels provided by different scan mode. Experimental results show that the suggested method can make higher accurate prediction and have a significant reduction of zero-order entropy in 3%-12%, as compared to CALIC with only one horizontal scan mode.