旋转纹理不变模型下的快速人脸匹配方法

Method of fast face matching based on rotation texture invariant model

为解决传统SIFT模型在人脸匹配时存在的相似区域误匹配问题并满足实时要求,提出一种RISIFTTF(rotation invariant SIFT with texture feature)模型下的快速人脸匹配方法。将RITF(rotation invariant texture feature)与SIFT融合,形成RISIFTTF模型,通过GPU对构建高斯差分金字塔和检测定位极值点等步骤并行加速。选取来自FERET和自采集的人脸库,对比优化前后的结果,对比结果表明,对于FERET库,配准率提高了8.23%-12.76%,加速比提高了5倍-6倍;对于自采集库,配准率提高了11.53%-14.21%,加速比提高了4倍-5倍,实现了对传统模型的优化与加速。

To solve the problem of the similar region mismatch and to meet the model, a fa-t face matching method based on RISIFTTF （ ro ta t io n invariant S IF T w i th texture feature） model was presented. The RISIFTTF model was formed which was the fusion of R IT F （ ro ta t io n invariant texture feature） and lel acceleration was used in some steps such as constructing the Gaussian difference pyramid as w e ltion of extreme points. The face database from F ERET and sel--collection were used to compare the results before and after opt-- mization. The results show th-t for the FERET face database, the registration rate is increased by 8. 23%-12. 76% and the speedup is improved by 5-6 times. F or self-collection,registration rate is increased by 11.increased by 4-5 times, achieving the acceleration and optimization of the traditional model.