基于光谱融合的手掌异常纹识别

摘要

针对现有掌部封闭型病理纹识别算法提取的线特征较少、 识别率较低的问题, 提出一种基于非下采样剪切波变换(NSST: Nonsubsample Shearlet Transform)域光谱融合的手掌异常纹识别算法.首先, 选取融合效果最佳的多光谱掌纹波段组合, 并在NSST域内进行多尺度、 多方向的分解;其次, 根据分解各层子带图像的特点设计融合规则进行相应系数矩阵的融合, 再通过NSST逆变换和形态学处理提取精细纹路特征;然后, 利用像素点的度特点寻找符合要求的闭合纹线回路;最后, 采用一种基于矩形度和偏心率等形状描述符的方法识别封闭型异常纹.实验结果表明, 该识别方法能提取丰富的掌纹线特征, 同时, 还可准确识别6种不同类型的封闭型病理纹, 识别率可达90%以上.%In order to solve the shortcoming of fewer extracted line featur e and lower recognition rates of pathologic palmprint recognition algorithm, we proposed a recognition algorithm of pathologic palmprint based on spectral fusio n in non-subsampled shearlet domain.Firstly, the best spectral fused combinati on of multispectral palmprint is selected.And it is decomposed to the multi-di rections, multi-scales in non-subsampled shearlet domain.Next, according to a ll levels characteristics of sub-bands images which had been decomposed, a new fusion rule is designed to fuse the corresponding coefficient matrices.The fine lines feature of palmprint can be obtained by the inverse transformation of the NSST(Nonsubsample Shearlet Transform) and the process of mathematical morpholog y.Then the satisfactory closed circuits are searched by degree feature of pixel s.Finally, we proposed a method of combining the shape descriptors based on rec tangle degree and eccentricity to recognize closed pathologic palmprint.Experimental results show that this algorithm can extract rich feature of the palmprint line, and can recognize six different types of closed patho logic palmprints accurately and the recognition rate is more than 90%.