分块Radon变换在低对比度掌纹特征提取中的应用

摘要

特征提取是低对比度掌纹识别的关键步骤.针对掌纹纹理特征明显的特点,本文提出了一种分块Radon变换的掌纹特征提取方法.该方法先对掌纹感兴趣区域进行一级小波分解去噪降维,接着对低频子图像进行分块以圈定局部主要纹理,最后把所有分块后的子图像进行70°～140°Radon变换,所获得的线积分组合在一起构成该图像的特征向量.运用UST掌纹图像库,对本文算法进行了测试.从识别率达到94％的实验结果看,此方法能够满足对采集图像无过多要求的认证系统的使用.%Feature extraction is a critical step for low contrast palmprint recognition. According to the characteristics with obvious palmprint texture features, a block Radon transform method is presented for palmprint feature extraction. Firstly, the method denoises region of interest of the palmprint and reduces dimension with the first-level wavelet decomposition. Then it blocks the low-frequency sub-image to delineate local primary textures. At last, after all the sub-block images are transformed during 70°~140° by Radon transform method, the line integrals obtained constitute feature vector for the image. The method was tested on the basis of UST palmprint image database. From the experimental results of 94% recognition rate, the method can satisfy the application for authentication systems without excessive demands for collection images.