The invention discloses a method for segmenting an ore image, comprising the following steps of: binaryzing an ore image on a conveyor belt by a self-adaptive threshold method; changing distance gray scale of a binaryzation image, keeping the black of a background pixel in the image unchanged, converting the gray value of a target pixel into the minimum distance with target boundary, ensuring that the center of each ore object erects like a peak and the pixel value at the adhered part of ores is lower than the pixel value of the ore center; integrally reducing all pixel values of distance images and enabling the adhered parts of ores to become the background and each object to be separated; removing the convex-concave at the peak of each object by reconstructing the distance images and enabling the peak to become a platform area as a seed area of the ores; and recording the edge of each seed area and growing the edges of the seed area in parallel to reach the boundary of each ore object. The invention can accurately and effectively segment the ore image.