Abstract: A license plate character segmentation algorithm, robust to poor images is presented. First, the non-uniform illumination correction, contrast enhancement and incline correction operations are introduced; then, the scale-adaptive cubic B-spline wavelet transform is applied to the outer-contour vertical distance of character regions for coarse segmentation; finally, character recognition feedback with template matching based on object occupancy rate is used for fine segmentation. Experimental results show that the proposed algorithm is robust in dealing with uniform-illuminated, contrast-poor, declining, dirt, character-touching and character-broken number plates.