Parallel Implementation of LBP Based Face Recognition on GPU Using OpenCL

Abstract

In this paper, we implement a novel parallelized approach of Local Binary Pattern (LBP) based face recognition algorithm on GPU. High performance rates have been achieved through maximizing the resource exploitation available in the GPU. The launch of GPU programming tools like Open source Computation Language (OpenCL) and (CUDA) have boosted the development of various applications on GPU. In this paper we implement a parallelized LBP algorithm on GPU using OpenCL programming tools. Programs developed under the OpenCL enable us to utilize GPU for general purpose computation with increased performance efficiency in terms of execution time. The experimental results based on the implementation on AMD 6500 GPU processor are observed to increase the computational performance of the system by to 30 folds in case of 1024×1024 images. The relative computational efficiency increases with increase in the size of the Image. This paper addresses several parallelization problems related to memory access and updating, divergent execution paths, understanding and realizing the OpenCL's concurrency and Execution models.