The invention discloses an endoscope image processing method. The method comprises the following steps of S1, using Gaussian filtering to remove a white noise generated during an image acquisition process; S2, calibrating a mapping relation f(r, c)-k{n} of an original image and a newly-constructed image so that a smallest circle bright spot circle center (r, c) in the original image is mapped to a rhomboid area (b1, b2, b3, b4) in the newly-constructed image, and acquiring an area block set k{n}; S3, according to a result of the step2, carrying out grid removing processing on subsequent images; S4, filtering the images acquired from the step3 again so that the images are smooth and continuous; and S5, storing and displaying the processed images. Wavelet transform and a threshold contraction method are used to preprocess collected image information, and through grid removing, gamma correction and filtering smooth processing, image quality is greatly increased.