The characteristics in fundus hemorrhage images are varying hemorrhage shapes and multiple interferences. Specific to these characteristics, a method based on three color channels in fundus hemorrhage images was proposed to improve the extraction accuracy of the hemorrhage areas and reduce the interferences caused by non-bleeding ones. Firstly, the relevant features of pixels in different color channels were analyzed by the statistical properties of the hemorrhage areas, and the extracting threshold was set in the method. Then, multi-scale top-hat transformation and vascular density feature were applied to locate vessels and macular for removing these disturbances. Finally, the logical relations of the hemorrhage, vessels and macular were computed to extract the hemorrhage areas while removing the interferences of vessels and macular. The proposed method realizes the automatic extraction of hemorrhagic areas and the simulation results show that the method can ensure the extraction accuracy with a high computational efficiency.