摘要
针对目前矢量量化技术存在的主要问题之一:图像边缘失真严重,本文讨论了一种应用神经网络的图像边缘保持矢量量化方法.它以Kohonen的自组织特征映射算法(SOFM)为基础,根据人的视觉系统对图像边缘的敏感性,先对整幅图像的边缘提取,再根据不同训练矢量(图像子块)的边缘特性,自适应地调整SOFM算法中的学习速率因子.本文中,图像的边缘提取及矢量量化,都是由神经网络实现的.实验结果表明,和单纯用神经网络直接进行矢量量化相比较,应用这种技术的图像编码在同一压缩比下译码图像的边缘质量有明显的提高.
To reduce the edge degradation of the coded image, an adaptive learning method of the edge preserving vector quantization (EPVQ) based on kohonen's self-organizing feature map neural network is presented. The computation is implemented by the neural network. Compared with direct image vector quantization coding, the experiment results show that the quality of the restored image is well improved at the same compression ratio.
出处
《模式识别与人工智能》
EI
CSCD
北大核心
1997年第3期265-270,共6页
Pattern Recognition and Artificial Intelligence
