3D palmprint identification combining blocked ST and PCA

Abstract

Three dimensional (3D) palmprint technologies have been widely studied as a method of human identification and recognition, because they offer unique merits over their 2D counterpart. To overcome the limitations of small sample sizes and low one-to-many identification speed, a novel method by combining the blocked surface type (ST) feature and principal component analysis (PCA) has been developed for 3D palmprint identification. This method adopts the histogram of blocked ST as an effective palmprint feature, thus reducing the subsequent computational complexity. The dimensionality of the data is reduced using the histogram method, and the 3D information is further compressed using PCA. A nearest neighbor classifier acts as the discrimination criterion for identifying a person. Experimental results using two databases demonstrate the effectiveness of the proposed method. Compared with other single-feature methods, the proposed approach overcomes the traditional problem of small sample sizes, reduces the computational complexity, and enables accurate, fast, and robust identification. Therefore, the proposed method is especially suitable for large-scale databases.