A Multi-threshold Approach to Segment Man-made Targets from Infrared Image by Discrete Stationary Wavelet Transform

In order to reduce the noise in the infrared image, gray-level histogram was smoothed by Bezier curve and Bezier histogram was obtained. One dimension stationary wavelet transform was implemented to the curvature curve of Bezier histogram. Position of peak values of curvature curve in wavelet domain were adjusted from "fine-to-coarse" at all scales. The gray level values, which were located in adjusted peak values at all scales, was considered as segmentation thresholds. The gray level values of valley between peaks were considered as quantity gray levels. Optimal segmentation scale was obtained by a cost criterion. The results of experiment show that a target can be segmented effectively from complex background in infrared image by new approach. The new approach is better than the segmentation methods, which were given by Hairong Qi and Jung Shiong Chang