Pulse Coupled Neural Network based anisotropic diffusion method for 1/f noise reduction
Deng Zhang, Toshi Hiro Nishimura · 2009
The effectiveness of anisotropic diffusion based denoising methods on both noise suppression and edge preservation have been demonstrated in many previous researches. However, the appearance of pulse noise liked spots in the denoised images due to a few high level noises in the noisy images becomes one of its limitations. This paper presents a pulse coupled neural network (PCNN) based anisotropic diffusion method to solve this problem at the same time of 1/f noise reduction in the pinned-type CMOS image sensors (CIS). Different from the traditional methods, pixels are respectively pre-processed by a median filter and a Lee filter according to the time matrix of the PCNN. Experimental results reveal that the pulse noise liked spots are eliminated by the proposed method. And finally, conclusions on the better performance on both 1/f noise reduction and edge and detail preservation are carried out compared with previous de-noising methods, i.e., median filter, Wiener filter, Lee filter, and traditional anisotropic diffusion based filter. Furthermore, the results will be applicable to the CIS manufacturing and also contribute denoising to still camera and video camera.