in 1979 Lee's paper, Lee Filter Digital Image enhancement and Noise Filtering by use of local Statistics, proposed the removal of additive noise based on partial information Sound, multiplicative noise and additive multiplicative mixed noise method, after careful study and coding, found that the method of removing additive noise is very good, with some of the current EPF algorithm similar to the edge retention function, and its operation is not complex, can be applied to similar to the grinding of the project. a simple algorithm is described below, for a n*m size grayscale image, with the pixel value at the representation (i,j) position, then the mean and local mean variance within the (2*n+1) * (2*m+1) window can be expressed as: and the result of additive de-noising is: among them: formula (4) in σ for the user input parameters. It's a simple process that smoothes the image but keeps the edges basically unaffected, such as the result:This excellent quality makes it possible to play a role in the image peeling. Let's look at how efficient this algorithm is. From the above calculation formula can be seen, its main calculation is the local mean and average distribution mean variance, the mean of the calculation of many optimization methods, typical such as cumulative integration chart. And on the average distribution of the optimization of mean variance, we recommend you see here: Http://fiji.sc/Integral_Image_Filters, the core of the demolition formula is: after this derivation, it can be seen that the local mean variance can also be achieved quickly by accumulating integral graphs, the result is that the efficiency of the program is independent of the local radius parameter, so the efficiency is very high. the above formula is for the grayscale image, for the common RGB color map, as long as the r/g/b three channels respectively to deal with is OK. with the above basis, after a personal groping, for the grinding application, the algorithm's two parameter radius is desirable: Max (src->width, src->height) * 0.02
Image denoising based on local mean-variance correlation information and its application in real-time skin beauty algorithm.
