Defogged image denoising method based on transmissivity

Abstract

A defogged image denoising method based on transmissivity comprises the steps: (1) obtaining a defogged image containing noise and a transmissivity image, (2) roughly estimating an overall noise variance, (3) according to the noise variance confirmed in the step (2), confirming and calculating several parameters, (4) according to the parameters, searching similar pixels, using the transmissivity as a criterion, pre-screening the pixels in the searching range, and removing points dissimilar to pixels to be denoised, (5) constructing a correction factor based on the transmissivity and weakening pixel weight polluted by the noise, (6) calculating a gradient mean value difference of a whole image and a searching window, inputting a built parameter optimization model, and conducting adaptive correction on denoising parameters in different areas, (7) calculating a similarity weighted mean value of all the pixels in the searching range to replace the pixels to be denoised, and (8) outputting a denoising result after all the pixels are processed. The defogged image denoising method has the advantages of being high in detail protection degree, good in denoising effect, low in complexity and the like.

Description

technical field [0001] The invention mainly relates to the technical field of image signal processing, in particular to a transmissivity-based dehazing image denoising method, which is applicable to non-local mean denoising, dark channel prior dehazing, parameter optimization modeling and the like. Background technique [0002] Under fog, haze and other weather conditions, the reflected light of objects in outdoor scenes is attenuated and scattered to varying degrees, resulting in serious quality degradation problems such as contrast drop and color shift in the collected images, which has become a threat to urban traffic, sea and air navigation safety. one of the main factors. [0003] Some practitioners proposed an image defogging algorithm based on dark channel prior, which significantly improves the visibility of images. However, the noise problem was not considered. Foggy images collected by low-cost cameras usually have a lot of noise, and the noise will be enhanced t...

AI Technical Summary

Problems solved by technology

[0012] 1. Denoising parameters need to be manually set;

[0013] 2. In step (2), it is necessary to operate on all pixels in the neighborhood of each to-be-denoised point, and the calculation cost of the algorithm is high and the running time is long;

[0014] 3. In step (3), the credibility and correctness of the given pixel similarity weights have not been considered;

[0015] 4. It does not take into account the characteristics that the dehazing image noise is mainly concentrated in bright areas such as the sky and dense fog, and the denoising results cannot maintain effective details

Some people also consider the interaction between the denoising strength, the noise variance and the comparison window, and model the weight attenuation coefficient h as h 2 =2β|N i |σ 2 , but still only determine the optimization parameters in the global sense of the image, and there are still varying degrees of local detail loss; someone else proposed a method for adaptively determining h based on the empirical standard deviation of the pixel block, although the edge is better protected fine detail, but falsely removes non-sequential, non-repeating patterns in the image

Others perform non-local average processing on the noisy image twice, and use the method noise calculated by the first processing result as a basis to correct and determine the similarity weight of pixels in the second processing, thereby improving the denoising effect, but must be at the expense of processing time

Images