Iris localization using parallel computing

Abstract

In this thesis, we have proposed a parallel iris localization technique by implementing canny edge detection in parallel on Graphical Processing Units(GPU) with the help of Compute Unified Device Architecture(CUDA) plateform. The output of canny edge detector which is binary image transfer from GPU/Device to CPU/Host and it is given to serial circular hough transform as input that locate the iris region from image. In this thesis, we follow the Wilde’s approach of iris recognition in which he used the edge detector, and Circular Hough Transform for detecting iris region from an eye image.We processed canny edge detection part of iris localization on GPU in parallel manner and Hough transform serially on CPU. In edge detection, we processed a number of pixels in parallel that execute on cores of GPU in block and thread manner, that reduces the execution time. The outcome of canny edge detector given to serial hough transform that locate the iris region from image. Then we compare the execution time of our parallel technique with existing serial one. In our case, execution time is reduced by 10 to 12 percent in comparison of serial approach. We use the 96 core NVidia GeForce GT 630 GPU for implementation.