Skin lesion image segmentation method based on diffusion difference

The invention relates to a skin lesion image segmentation method based on diffusion difference. The method comprises the following steps: S1, acquiring a medical image data set containing multiple skin lesion types; s2, a training set X is obtained after image preprocessing; s3, building a diffusion model, and training the diffusion model by using images with benign and malignant semantic information in X to obtain a trained model, S4, inputting an unlabeled skin lesion image x '0 into the trained model, and obtaining a plurality of different segmentation result sets Y of x' 0 by means of the multi-output capability of the trained model, and the uncertainty of the segmentation result is measured by means of the thought of generalized energy distance. And S5, in combination with a conditional random field algorithm, further optimizing the Y to obtain a more accurate segmentation result. According to the method, hidden semantic information is mined by learning the difference between different noises, semantic s