The embodiment of the invention provides an image segmentation method and device based on confidence propagation, and the method comprises the steps: carrying out the wavelet transformation of a to-be-segmented image, and obtaining a feature vector of each scale; for any scale, performing intra-scale iterative segmentation based on a confidence propagation algorithm, and calculating global energycorresponding to each iterative segmentation result until a preset intra-scale iterative condition is reached, wherein the global energy comprises marking field energy and characteristic field energyobtained based on an iterative segmentation result and a characteristic vector; taking the iterative segmentation result corresponding to the minimum global energy as the final segmentation result ofthe scale, taking the final segmentation result of the scale as the initial segmentation result of the next scale, and performing inter-scale iterative segmentation until the final segmentation resultof the finest scale is obtained; and taking the final segmentation result of the finest scale as the final segmentation result of the to-be-segmented image. According to the method and the device provided by the embodiment of the invention, globally optimal image segmentation is realized, and the accuracy of image segmentation is improved.