Adaptive noise reduction method using variable window size based on region analysis

This paper proposes a novel denoising method which uses a variable window size for noise filter depending on the results of local region analysis. The proposed method classifies local regions of a given noisy image as detail or smooth regions using a new parameter which is calculated by the ratio of estimated noise variance to local variance. The new parameter, a detail descriptor, represents the amount of image details in a local region. Then, the proposed method adaptively applies a noise reduction filter to remove image noise while preserving image details by regulating the filtering window size depending on the detail descriptor.