基于神经网络的图像KL变换方法的改进

KL变换作为最优变换在图像压缩中很有应用潜力.传统的基于神经网络的图像KL变换方法存在一些不足.本文提出了一种基于神经网络的图像KL变换的改进方法.该方法的特点是:通过对图像进行行列两次分割得到两组学习样本,分别对两个神经网络进行训练,用训练好的两个网络对原图像进行二次KL变换.对新方法进行仿真,结果表明所提的方法图像压缩效果较好,有效的消除了变换对于分割方向性的依赖.

Improving Karhunen -Loeve Transform of Images Based on Neural Networks

KL transform, as the optimal transform, can be applied to image compression potentially, There are some shortcomings for the conventional image KL transforms based on neural networks. This paper introduces an improved method for image KL transform based on neural networks. In the method, two KL transformation matrices respectively representing row direction and column direction of image are obtained by training neural networks, and used to complete a two times of KL transform. The simulative result shows that the improved method can eliminate the dependence of directions in transformation and give a stable good result.