Region-of-interest-based algorithm for automatic target detection in infrared images

We present a region-of-interest-based segmentation (ROI-S) algorithm and apply it for automatic target detection. The proposed algorithm requires no templates or a priori knowledge of the targets. An automatic ROI extraction approach based on localized texture and statistical features is used to locate targets in an IR scene without any prior knowledge of their type, exact size, and orientation. Two locally adaptive histogram-based segmentation techniques are applied to extract the target signature. The Bayes decision rule is applied for a bimodal histogram while entropic correlation is used for all other cases. Geometric and statistical features are automatically extracted for each suspected ROI. We suggest a unique variance-based metric for discriminating targets from clutter and for evaluating the probability of correct detection. The proposed system is successfully tested on several hundred single-frame IR images that contain multiple examples of military vehicles, with various sizes and brightness levels and in various background scenes and orientations. A high probability of correct detection (greater than 90%) with a low false alarm rate is achieved.