基于仿射尺度不变特征变换的掌纹识别

Palmprint recognition based on affine scale invariant feature transform

相比传统接触式采集方法,非接触采集是目前掌纹识别的趋势和主流,但其低约束性可能导致人手和图像传感器平面不平行,从而使掌纹产生仿射变形。传统的尺度不变特征变换(scale invariant feature transform,SIFT)对此识别效果不佳。针对这个问题,提出一种改进方案,即基于仿射尺度不变特征变换(affine scale invariant feature transform,ASIFT)的掌纹识别方法,建立了变形掌纹的仿射模型,模拟了相机光轴的经度角和纬度角,在仿射空间内提取图像特征。通过基于实际环境所建立的掌纹库———SUT图库验证算法性能,与SIFT算法及目前典型的掌纹识别方法进行对比。结果表明,ASIFT方法具备良好的抗掌纹仿射变形性能,等误率仅为0.6%,证明了该方法能够成功解决掌纹变形问题,鲁棒性和稳定性强,具备优越性。

Compared with traditional contact collection method, eontaettess acqmsmo- between of palmprint recognition. However, this method may lead to affine deformation caused by non-parallelity palm plane and sensor plane. In order to improve the limitation of scale invariant feature transform （SIFT） about this problem, a better palmprint recognition method based on affine scale invariant feature transform （ASIFT） is pro- posed. Firstly, an affine model of the deformed palmprint is established, then the latitude and longitude angles of the camera axis are simulated, and the image features in the affine space are extracted. Based on the practical applica- tion environment, the SUT panflprint database is established for the performance tests. Compared with the SIFT and other typical palmprint recognition methods, the experimental results show that AS1FT has good performance in resis- ting the affine deformation of palmprint, the equal error rate （EER） is 0.6%. In conclusion, the proposed algorithm can successfully solve the deformation problem of palmprint, and has superiority, and strong robustness and stability.