This paper presents an implementation of local binary pattern algorithm for face detection using different approaches and a comparative study between them. The comparison is based on both the processing time and accuracy for two famous approaches; feature scaling and image scaling. The two approaches are implemented on the GPU and the pros and cons of each approach are addressed. The results of this paper focus on two aspects; accuracy and number of processed frames per second. It is found that the GPU implementations can process up to 50 FPS (frames per second) for feature scaling. It is also able to process up to 45 FPS for image scaling. In both cases, the GPU-based implementation has maintained the same level of face detection accuracy. In addition, it was found that image scaling almost has eliminated false detections.

GPU, local binary pattern.