Improved MAP Estimation of Variance Through Arbitrary Windows For Wavelet Denoising

For the past two decades wavelet transform has been a promising tool in image processing. Here a novel method for Gaussian noise removal in images is proposed by estimating the signal variance from noisy environment. As wavelet coefficients are correlated with each other the size of the window considered for estimating variance becomes a critical factor. Previously Maximum Likelihood (ML) and Maximum A Posteriori (MAP) methods were used with fixed windows. Here instead of fixed window, arbitrary shaped windows are used. Testing the similarity of variance with an adaptive threshold generates these windows. For this arbitrary window a modified maximum a posteriori estimate for signal variance is proposed. Finally the denoised coefficients were estimated through LMMSE estimate. The simulation results show improvement performance over the state of art wavelet denoising procedures in PSNR measures with good visual quality.