Aiming at problem that traditional defect detection of paper surface algorithm is easily to be interfered by complex backgrounds

a method for surface defects detecting based on relative homogeneity between classes is proposed. This algorithm assumes that the foreground has gray uniformity

and uses the criterion of the maximum between-class variance（ Otsu） and discriminate criterion of foreground gray uniformity to deduce new threshold criteria. Establish a threshold model with relative homogeneity between classes. In the special case of uneven background

information entropy combines with the threshold model to obtain the optimal threshold. The experimental results show that compared with the existing algorithms

this algorithm can more accurately segment the defects which has smaller area and with background interference. Compared with the existing algorithm

running time of single frame is shortened. The algorithm improves accuracy and efficiency of paper detection.