Parallelized and vectorized implementation of DCT denoising with FMA instructions
Yuki Kawasaki, Yoshihiro Maeda, Norishige Fukushima · 2018 International Workshop on Advanced Image Technology (IWAIT) · 2018
DCT denoising is a denoising technique, which filtering an image in the frequency domain. The DCT denoising is known as an excellent method in a balance between processing time and denoising accuracy. In this paper, we implement the DCT denoising for further improving the computational cost to use efficient implementation of DCT, named AAN (Arai-Agui-Nakajima's methods). Also, we utilize FMA (fused multiply-add) instructions for the AAN-based DCT for accelerations. In the experiments, we compare the proposed DCT algorithm with DCT denoising based on Chen's algorithm, which is a normal fast DCT algorithm. The experimental results show that the proposed method is superior to the conventional method.