Single Image Noise Estimation Based on Image Segmentation and Scatter Statistics of Noise

Abstract

Single image noise estimation is important for evaluating image and optical imaging system. A no-reference image noise estimation approach based on noise scatter statistics and accumulation of RGB three channels is proposed. An image is filtered by contrast sensitivity function (CSF) to match the human visual system, and the image is segmented by watershed method into sub-image patches with different contents, different patches are achieved free noise estimation using affine reconstruction. Then the pairs of intensity and noise are calculated for patches, and a scatter diagram for RGB three channels of intensity and noise is obtained. Through noise statistics and accumulation , the noise level estimation is achieved. The general database is applied to test the proposed method and other methods in the experiment. Experimental results indicate that the proposed method achieves the same evaluation both objective and subjective.