基于Kohonen神经网络的图像超分辨率处理算法  

Image Super Resolution Processing Algorithm Based on Kohonen Neural Network

作　　者：陶洪久[1] 徐洪波[2] 柳健[2] 田金文[2] 

机构地区：[1]武汉理工大学物流工程系,武汉430063 [2]华中科技大学图像信息处理和智能控制教育部重点实验室,武汉430073

出　　处：《武汉理工大学学报（交通科学与工程版）》2002年第4期474-477,共4页Journal of Wuhan University of Technology（Transportation Science & Engineering）

基　　金：北京大学国家重点实验室基金 (2 0 0 1 );国防科工委"十五"民用航天预研会基金资助 (2 0 0 1 )

摘　　要：提出一种小波变换、Bézier曲面插值和神经网络相结合的方法 ,并将该方法应用于图像处理中 ,对图像进行超分辨率处理 ,以提高图像质量 .首先将原图像作为低通部分 ,然后对小波分解后的相应高频子带进行 Bézier曲面插值 ,再对插值后的高频子带分别进行神经网络学习重建恢复各高频子带图像 ,以近似更高频的细节 ,最后通过小波逆变换获取比原图分辨率更高的图像 .模拟实验结果表明 :与小波插值方法相比 。The paper proposes an algorithm used in image processing, combining wavelet transform and Bézier surface interpolation with Kohonen neural network. The image is implemented by super resolution processing, which can improve the image quality. First, the original image is regarded as low frequency part. Then Bézier surface interpolation are implemented to the high frequency sub band images after wavelet transform. Then, the high frequency sub band images after interpolation are implemented respectively by Kohonen neural network algorithm to reconstruct three high frequency sub band images, which is approximate to details of higher frequency. Finally, an image with higher resolution is obtained by inverse wavelet transform. Experiments show that the proposed algorithm can improve image quality much more in comparison wich wavelet transform and Bézier surface interpolation algorithm.

关 键 词：算法 小波变换 BÉZIER曲面 KOHONEN神经网络 图像处理 超分辨率 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]