LST变换和手部几何特征融合的模糊掌纹识别

Blurred palm-print recognition based on fusion of Laplacian smoothing transform and geometric features of hand

为了解决非接触采集时的离焦状况容易导致掌纹图像出现模糊现象,从而造成识别系统的性能降低的问题,在建立模糊模型并分析模糊现象机理的基础上,提出了一种新的解决方案。使用拉普拉斯平滑变换(laplacian smoothing transform,LST)提取模糊掌纹的低频系数作为稳定特征,提取手部几何特征即手指相对长度和宽度作为特征向量,将LST特征和几何特征进行融合,最后利用特征向量之间的欧式距离进行匹配和分类。在自建的SUT-D模糊掌纹图库上的测试结果表明该算法等误率可达7.01%,与融合之前及其他典型识别方法相比,等误率最高可降低13.11%,显示出该算法具备有效性,为解决模糊掌纹的识别问题提供了一条可行途径。

The defocus status caused by non-contact collection for palm-print will lead to image blur. To improve recognition performance of the identification system, a novel solution is proposed based on establishing the blurred model and analyzing blur mechanism. In this paper, the Laplacian smoothing transform （LST） is employed to extract the low-frequency coefficients of the blurred palm-print as the stable features, the hand geometric features, namely the relative lengths and widths of the fingers are also extracted as the feature vectors, the LST features and geometric features are fused to constitute the new vectors. Finally, Euclidean distance between the vectors is used for matching and classification. The experiments based on the self-made SUT-D blurred palm-print database show that the equal error rate （EER） of the proposed algorithm can achieve 7.01% ,which could be maximally reduced by 13.11% compared with no fusion and other typical identification methods. It demonstrates that the algorithm is an effective and superior approach which can solve the problem of blurred palm-print recognition.