To solve the problem of stacked papers counting, the method combined minimum curvature with peak detection is presented. First, the profiles of the stacked papers along the vertical orientation of the paper are extracted. The curvatures of the profiles are calculated after processing the images by Gaussian filtering. Second, the centers of the stacked papers need to be detected. As the profile of the sheet is like the shape of Gaussian. So, the local minimums of curvatures indicate the center positions of stacked papers. The width of the region where the curvature is negative represented the thickness of the paper. Then the center positions of stacked papers can be corrected by judging the distance between the two adjacent center points and the gray features. Finally, the counting result can be acquired by using peak detection from the ultimate ridge line result of the stack papers. The algorithm proposed in this paper can accurately detect the abnormal paper by fusing the gray features of the stacked papers and the distance of adjacent paper. The experimental results show that the error rate of the method is less than 0.01% for the paper with the thickness between 0.05~0.2 mm.