Abstract: To alleviate the limitations that the existing palmprint recognition methods are time-consuming, and their robustness to the variations of orientation, position and illumination in capturing palmprint images is insufficient, this paper describes a new palmprint recognition method based on Speeded Up Robust Features（SURF） and fuzzy reasoning. The proposed method consists of two phases: enrolling phase and identification phase. In the enrolling phase, keypoints are detected with SURF from the enrolled palmprint images and mutually matched between training samples, and then for each enrolled user, the keypoints which matching frequencies exceed 50% are selected. For each selected keypoint, the matching rate, the mean and variance of the location coordinates and the mean of SURF descriptors in the enrolled palmprint images, the maximal distance between the mean and the SURF descriptors are determined and regarded as the user’s palmprint patterns stored in a pattern library. In the identification phase, it utilizes SURF to detect keypoints from the palmprint image to be identified, and then the fuzzy matching degrees for all palmprint patterns in the library are calculated. A fussy reasoning approach is developed for matching. Experimental results demonstrate that the proposed method yields a better performance in terms of the correct classification percentages compared with the recent palmprint recognition algorithms. It is also shown that the proposed approach is robust to the variations of orientation, position and illumination, and yields observably low computational cost.

Key words: Speeded Up Robust Features（SURF）, point matching, fuzzy reasoning, palmprint recognition

摘要： 针对现有掌纹识别算法对掌纹图像在采集过程中的位置、方向、亮度变化缺乏足够的鲁棒性，而且计算复杂度较高的问题，提出了一种基于SURF描述字的掌纹识别算法。算法分为训练与识别两个过程，在训练过程中，提取属于同一类所有训练样本的SURF描述字进行互配，然后计算训练样本中互配频次超过该类样本数的1/2的每个关键点的匹配率及其在匹配训练样本中坐标的均值与方差以及SURF描述字均值、SURF描述字与均值的最大欧氏距离组成类别数据库。在掌纹识别过程，基于SURF提取待识别掌纹图像的关键点，确定关键点的SURF描述字与其位置坐标，然后，计算类别数据库中每个类别的每个关键点与待识别掌纹图像所有关键点模糊匹配度的最大值作为该关键点的模糊匹配度，最后基于模糊推理实现掌纹识别。实验结果表明该算法对掌纹图像的旋转、尺度和亮度的变化具有较好的鲁棒性，具有稳健和高精度的特性，并且识别过程计算成本较低，满足了实时性应用的要求。

关键词: SURF算法, 关键点匹配, 模糊推理, 掌纹识别