Pavement image enhancement based on scale evaluation using nonsubsampled contourlet coefficients

This paper describes a scale evaluation method using nonsubsampled contourlet transform and its application in pavement image enhancement for crack detection. Crack in some scales is much more visible than in others, so a method for scale evaluation is given, and different gains are delivered to each scale for enhancement after scale evaluation. In the first step, noise threshold is computed by noise estimation. And then, sub-groups with 64times64 pixels are divided from the full image at each scale, and group direction variances of these sub-groups are computed for scale evaluation. At last, enhancing process at different scales are taken with gains obtained from scale evaluation. Experiment results show a promising use of the presented method for pavement image enhancement.