Abstract: This paper deals with the challenging task of reconstructing a full three-color representation of color images captured by a single image sensor – CCD or CMOS, which samples the image according to a color filter array (CFA). The output images feature a certain kind of pattern in which every pixel has only single-color information. The process of modifying the raw image sensor data to obtain full RGB color information at each pixel is called demosaicing or CFA interpolation. Seven popular algorithms CFA interpolation are regarded. The results of timing performance analysis of their sequential implementations are presented and discussed. The performance speedup analysis and scalability tests with multithreading and CUDA-based GPU implementations are analyzed.
Keywords: Bayer pattern, CFA interpolation, GPU, Image sensors, Multithreading, Parallel computations, Performance and speedup analysis.