In consideration of the problem of over-segmentation or under-segmentation in high-resolution remote sensing image segmentation that the noise leads to, a high-resolution image segmentation method with phase consistency and watershed transformation is proposed. Firstly, the phase-consistent model method with spectral similarity is adopted to obtain the edge response amplitude, and then the automatic marker watershed algorithm is adopted to segment the image. Based on the multiple restrictions for the features, such as spatial position, shape and area of adjacent segmentation objects, adjacent segmentation object merging cost function model is proposed to optimize the segmentation result and obtain the final segmentation result. Experimental images of the typical area are selected for the segmentation experiment. The effectiveness of the proposed method is verified by visual evaluation, supervision evaluation and comparison with the typical segmentation methods.