Multiple feature fusion for unconstrained palm print authentication

Over the last decade, palm print recognition has emerged as the strongest technology for human authentication in many aspects. To carry out an effective recognition, this paper presents a feature level fusion of block-wise scale invariant feature transform and texture code co-occurrence matrix based features. Initially, an attempt to access the quality of extracted region of interest image is made. This is followed by application of fractional differential mask resulting in improvement of textural detail. In order to select the most discriminate palm features, a feature transformation algorithm inspired by subspace learning is employed. It led to reduction in computation time and feature dimensions, along with higher level of performance. A trained support vector machine utilizes the selected features to determine whether image belongs to genuine or imposter class. Comparative experimental analysis described in this paper indicates customarily outperforming results than competing methods and validate efficacy of proposed approach.