high performance computing on graphics processing units: hgpu.org

Iris recognition is quite a computation intensive task with huge amounts of pixel processing. After the image acquisition of the eye, Iris recognition is basically divided into Iris localization, Feature Extraction and Matching steps. Each of these tasks involves a lot of processing. It thus becomes essential to improve the performance of each step to gain an overall increase in performance. The localization step is of utmost importance since it finds out the essential region over which further steps of Iris Recognition are to be performed. It thus decreases the amount of computation that will be needed in the subsequent steps. In this thesis an effort has been made to improve the performance of Iris localization by the use of parallel computing techniques. Recently the General Purpose Graphics Processing Units(GPUs) have come to be very popular in solving complex computational tasks. In order to achieve a speedup in the localization step, the Compute Unified Device Architecture(CUDA) platform released by NVIDIA corporation has been used. Hough Transform for circles has been used to perform the localization step since it has the ability to handle noisy data very efficiently. The edge image has been obtained using the popular canny edge detector and it serves as the input for the Hough Transformation step. Since the image data as well as the edge detecting mechanism may not be perfect, the Hough transform method carries out a voting mechanism over the image objects, in order to deal with imperfections like noisy data. Parallelism is employed in the Hough transformation step, when for each possible value of the radius a large number of circles have to be generated in the parameter space, and this task is taken over by parallel blocks and threads, which substantially improves the computation time required to identify the circular contours in the image space.