Image noise estimation using color information

Staelin, Carl ; Nachlieli, Hila

HP Development Company

关键词: noise estimation;    denoising;    image processing;   

RP-ID  :  HPL-2005-2R1

学科分类：计算机科学（综合）

美国|英语

来源: HP Labs

【 摘 要 】

Imaging products are continuously increasing their optical resolution, resulting in higher perceived noise levels due to increased sensitivity to noise and graininess in the original image and to the decreased signal-to-noise ratio in the smaller pixel sensors. Denoising images requires accurate noise estimates; too low an estimate and too much noise remains in the image, too high an estimate and too many details are erased from the image. In addition, the noise level is often not constant over natural images, particularly images from grainy negatives. We present a novel noise estimation algorithm for scanned images, based on the statistical prior that the color channel gradients are highly correlated in natural noise-free images, extend it to provide local noise estimates, and then extend a denoising algorithm to utilize the spatially variant noise estimates for improved denoising while retaining details in the original image. 14 Pages