Image smoothing based on local image models

On the basis of an algorithm which identifies the neighborhood of a pixel as constant, oriented, or irregular based on the spatial relationship among the pixels inside this neighborhood, the use of different smoothing schemes according to the neighborhood surrounding each pixel is proposed. This approach is applied to images corrupted with additive Gaussian noise and impulse speckle noise. The flat regions of an image are smoothed sufficiently, while linear structures are enhanced.<>