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

摘要

The present invention reduces noise in digital photographic images and is based on the assumption that we can break down the image into two types of regions: the smooth regions and corner regions. The smooth regions are areas of the picture where one finds no specific detail is such a blue sky. The corner regions are regions containing specific details, such as corners and colored regions (such as grass). The method presented here reduced in smooth regions using a mathematical technique of image blur based on the regression of squares. The image blur does not affect the sharpness of the image because it has no precise details in smooth regions. The corner regions are not affected to maintain their sharpness, however, we notice less noise in these areas than in smooth regions. The method operates on the luminance component of a digital image as follows: 1). From a matrix of pixels representing an image, selecting a set of types of neighboring pixels. For example, four types of square neighborhood are selected: 21 x 21, 11 x 11, 5 x 5, 1 x 2 1.) For each pixel of the image (auqeul Referring as a target pixel), have been all neighborhoods (including all types of neighborhood selected) which contain that pixel. For each target pixel, there are neighborhoods 588 of this kind (441 to 21 x 21 121 11 x 11 25 to 5 x 5, and one size 1 x 1). 3) For each vicinity of the target pixel, a linear regression of least squares is computer processed to correspond to a plane of the code values ​​of the pixels in the neighborhood. (The image can be considered as a mathematical surface, and the value of reconstruction pixel code as a function of the two coordinated pixels. The method approximates the true surface of the image by means of a planar surface ) this gives a good fit (measured by X 2 which is the sum of the normalized square errors for