Computational techniques involving noise filtering on two-dimensional image arrays are developed based on their local means and variances. These algorithms do not require the use of any kind of transform. The noise filtering techniques such as Kalman filtering methods and transform domain methods require extensive image modelling and produce filtered image with considerable contrast loss.
Recently, several techniques of noise smoothing using the minimum mean square error(LMMSE) estimator were proposed in image statistics based on a nonstationary mean and nonstationary variance(NMNV) image model. These techniques assumed that each pixel is uncorrelated random process for removal of computational complexity.
This thesis derives the LMMSE filter for noise reduction of image with the consideration of the correlation between pixels. The filter is able to adapt itself to the image models with different types of noise such as multiplicative noise and film grain noise as well as additive noise. Examples on images containing 512×512 pixels are given and the processing results are compared with existing methods of filtering.
