In order to further improve the de-noising ability of conventional polynomial matching algorithms, an improved polynomial matching filtering algorithm based on edge protection is proposed, aiming at the details such as image edge / texture easily blurred by conventional polynomial matching algorithms. Based on the conventional polynomial matching algorithm, it improves the selection method of filtering window, extracts adaptive sliding filtering window along the direction of image texture direction, selects the window with minimum matching error for matching filtering and takes it as the final output result. Then the grayscale image and the actual CT image were tested respectively. The data verification shows that this method can maintain edge / texture information while effectively suppressing the quantization of noise, and the peak signal-to-noise ratio is increased more than 80%, and the root mean square error is reduced more than 80%. Compared with the conventional polynomial filtering method, median filtering method, bilateral filtering method, edge preserving filtering method, BM3D and DnCNN method, the improved method can effectively improve the image visual effect, meet the requirements of image application, and has a good application prospect.