Infrared Dim-Small Target Detection Based on NSCT and Three-layer Window Local Contrast

When using a sliding window to calculate local contrast, when the sliding window size is larger than the target size in the original image, it will cause a "swelling effect" that causes the target to be missed. In order to solve the above problems, this paper proposes an infrared weak target detection algorithm based on non-subsampled contour wave transform and local contrast of three-layer window. According to the global sparsity of the target on the infrared image, the non-subsampled contour wave transform is introduced to decompose the image into low-frequency and high-frequency sub-graphs, and the differential image of high-frequency and low-frequency subgraphs is constructed. Guided filtering can effectively enhance the signal strength of the target and increase the gray difference between the target area and the background neighborhood, and then calculate the local contrast with the three-layer sliding window for background suppression and target enhancement, and then construct a confidence map. In order to test the effectiveness of the proposed method, six groups of open-source infrared sequence images were selected for comparative experiments, each group of sequences included 30 frames of images, with different backgrounds and large differences, the experimental results showed that the algorithm effectively avoided the problem of target missed detection caused by the "expansion effect", and the ROC curve, PR curve and AUC value were used to evaluate the experimental results, and compared with the existing 8 algorithms, the proposed method in the ROC curve was in sequence 1, sequence 2 and sequence 6. A higher detection rate was always maintained at the same false alarm rate, and the AUC value was the highest of all methods, and the AUC value was also the second best value in the remaining sequences 3, 4, and 5. Similarly, in the PR curve, the proposed method maintains the highest precision under the same recall rate in sequence 2 and sequence 3, with AUC values of 0.9309 and 0.9506, which have good improvements in background suppression, target enhancement and accuracy.