摘要
Gabor变换是掌纹识别中提取纹理特征的一个重要工具,但其性能易受图像的变化以及不均衡噪声等因素影响,因此提出了一种基于Gabor局部相对特征的掌纹识别算法。该算法对原始图像进行微尺度不变Gabor滤波;结合分形学的思想,将滤波后的图像分成大小相等的子域,每个子域又分成多个相同的子块,计算每个子块与它所在子域的相对方差,将所有子块的相对方差排列组成表征图像的特征向量进行识别。该算法将微尺度不变与局部相对特性统一,所提取的特征对各种变化有很强的鲁棒性,提高了识别精度和效率。实验使用北京交通大学BJTU_PalmprintDB证明该算法的有效性。
Gabor transform is an important tool for texture analysis in palmprint recognition. However, it is sensitive to the variations and uneven noises. A novel method is proposed to extract Gabor local relative features for palmprint recognition. Micro-scale invariant Gabor filters are designed to convolute with the original images; and inspired by the fractals, Gabor filtered images are divided into partitions. The relative derivations of all partitions are calculated to compare the similarity between ranges and their resided domains, which composite the feature vectors for image representation. The algorithm combines micro-scale invariant and local relativity for robust features extraction, and therefore, the recognition performance is improved. Identification experiments are executed on BJTU_PalmprintDB to test the effectiveness of the proposed method.
出处
《计算机工程与应用》
CSCD
2012年第15期34-38,48,共6页
Computer Engineering and Applications
基金
国家自然科学基金(No.60472033
No.60672062)
内蒙古自治区高等学校科学研究项目(No.NJ10057)
内蒙古农业大学创新团队项目(No.NDPYTD2010-9)
关键词
掌纹识别
GABOR变换
相对特征
纹理分析
palmprint recognition
Gabor transform
relative features
texture analysis
