A COMPUTER BASED DIGITAL IMAGE NOISE REDUCTION METHOD BASED ON OVERLAPPING PLANAR APPROXIMATION

摘要

The present invention reduces noise in digital photographic images based on the assumption that images may be decomposed into two types of regions, smooth regions and edge regions. Smooth regions are areas of the image lacking any sharp detail, such as blue sky. Edge regions are regions containing sharp detail, such as edges and textured regions (such as grass). The present method reduces noise in the smooth regions by a mathematical blurring technique based on least squares regression. The blurring does not degrade the sharpness of the image, because there are no sharp details in the smooth regions. Edge regions are left undisturbed to maintain sharpness, but the noise is less noticeable in those regions than in the smooth regions. The method operates upon the luminance component of a digital image as follows: 1) From a matrix of pixels representing an image a set of neighborhood pixel types is cheson. For example, 4 square neighborhood types are chosen: 21 x 21, 11 x 11, 5 x 5, 1 x 1. 2) For each pixel in the image (referred to as the target pixel), all neighborhoods (of all chosen neighborhood types) which contain that pixel are considered. For each target pixel there are 588 such neighborhoods (441 of size 21 x 21, 121 of size 11 x 11, 25 of size 5 x 5, and 1 of size 1 x 1). 3) For each neighborhood of the target pixel, a linear least squares regression is computed to fit a plane to the code values of the pixels in the neighborhood. (The image may be considered a mathematical surface, plotting pixel code value as a function of the two pixel coordinates. The method approximates the true image surface with a planar surface) this results in a goodness of fit (measured by X2, which is a sum of squared errors normalized for the standard deviation at each pixel), and the least squares estimate for the code value at the target pixel. 4) A ''noise-reduce'' code value for the target pixel is computed as a normalized, weighted sum of the least squares estimates for the target pixel, summing over all neighborhoods containing the target pixel. The weights in the sum are functions of the goodness of fit (X2) and the the neighborhood type (21 x 21, 11 x 11, 5 x 5, or 1 x 1).