Abstract: JPEG-LS standard performs lossless and near-lossless high quality compression for still images.Its predictive coding is based on simple median edge detection scheme，but the redundancy is not well reduced.This paper explores the continuity property and the relativity of local pixel texture.It studies the selection of reference pixels，construction of texture-based nonlinear classification predictor，and design of prediction parameters.The gradient detection capability is enhanced，and a new four-order classification predictor is proposed.It is confirmed that the prediction performance is effectively promoted with low operation complexity.

摘要： JPEG-LS标准实现对静止图像的无损压缩以及近无损高保真压缩，其预测编码仅采用简单的中值边缘检测法。将充分利用邻域像素纹理的连续性与相关性，研究参考像素的选取、基于纹理信息的非线性分类预测器的构建与预测器参数的设计，增强梯度检测能力，提出新的四阶分类预测器。实验证明，该算法在低运算复杂度的前提下，有效提高了预测编码的性能。