02-02-2015, 06:11 PM
[/size][/font][font=Times New Roman][size=medium]
Posts: 14,118
Threads: 61
Joined: Oct 2014
Recognition of the sclera vein has been shown to be a promising method for human identification. However, their match speed is slow, which could affect their application for real-time applications. To improve the efficacy of adaptation, a new method of recognition of the parallel sclerotic vein was proposed using a two-stage parallel approach for recording and adaptation. First, we design a rotation-based and Y-scale-invariant feature extraction method to efficiently eliminate the most improbable matches. Second, we developed a polarized sclera descriptor structure to incorporate mask information to reduce the memory cost of the GPU. Third, we designed a method of comparison of two stages of thickness to end. Finally, we developed a mapping scheme to assign the subtasks to GPU processing units. Experimental results show that our proposed method can achieve a dramatic improvement in processing speed without compromising accuracy of recognition.
Related video is here :