To protect image features such as edge and texture information,an FG -SVM denoising method based on fractional integral operator and image gradient is designed.First,improve PCNN to enhance the accu racy of noise detection,and detect the position of noise pixels on the noise image,and generate mark image by marking the correspo nding position of noise pixels and signal pixels as 1and 0,respectively.Second,according to the mark image,in every 5×5region with signal pixel in the central,construct training samples using pixels grey information around the central pixel point,fractional -integral operators,and gradient information,then, train SVM with all training samples.Third,set up test sample in every 5×5reg ion with noise pixel in the central,and use it as the input of the trained SVM to estimate the pixel value of the central.Finally,th e estimate values of the SVM were used to replace noise pixel values,get the final denoised image.The simulation results show that it can obtain the best denoising effect when the fractional order is equal to 1.7±0.1.When denoising the images of Lena,Pepper and Camer. with noise concentration of 1%,the PSNR of FG-SVM is increased by [4.19,1.60,3.64] dB than MPCNN,FG-SVM produced visua lly pleasing denoising image with clear edge information.