图像的小波系数神经网络预测编码

提出一种图像的小波系数预测编码方法。在该方法中，原始图像被分解为一个小尺寸的低频图像和一系列不同方向的多分辨率细节子带图像（小波系数图像），由于不同分辨率的小波系数图像具有相似性，故可用低分辨纺小波系数来预测同一方向相邻较高分辨率的小波系数。采用Ｂ－Ｐ神经网络作为非线性预测器有效地实现了这种预测，结果表明，利用这种编码算法可获得高图像压缩比。

IMAGE CODING BASED ON WAVELET COEFFICIENT PREDICTION USING NEURAL NETWORKS

?An image-coding scheme based on wavelet coefficient prediction is propoed.The image is decomposed into a low frequency subimage with small she and a set ofmultiresolution detail subimages (i. e. wavelet coefficients) along different orientations (horizontal,vertical and diagonal). There exists similarity among the detail subhages, therefore, the waveletcoeffidents in the higher resolution subimage can be predicted from those in the lower resolu-tion subimage along the same orientation. As the nonlinear predictors, the B-P neural networks are ed for the ptalictfon. Some examples show that the coding scheme leads to ahigh companion ratio for 512 x 512 still images.