The invention relates to a curvature variation based wavelet image denoising algorithm. The curvature variation based wavelet image denoising algorithm is characterized by comprising the steps of 1, algorithm description, namely, performing wavelet transformation for an input image to be denoised, introducing a horizontal set curvature as a correction factor into a variation model, and creating the curvature variation based wavelet image denoising algorithm; 2, algorithm verification, namely, the first item for the curvature variation model is a dispersion item in the image smoothing process, and the second item of the curvature variation model is designed to be the control function of the image structure, to maintain the integral structure of the image; 3, algorithm simulation, namely, performing the simulation algorithm of MATLAB software, and analyzing the timeliness and complexity of the algorithm through the simulation result. With the adoption of the algorithm, the processed image can be clear and close to the original image; the signal to noise of the denoised image is increased by about 15dB by being compared with that of a TV model; the classic wavelet threshold denoising algorithm is increased by about 25dB, and moreover, the definition is greatly improved.