Research On Methods Of Infrared Image Target Detection

Precision-guided weapons become more and more important in modern warfare. Infrared guidance technology based on forward-looking infrared imaging is a main development direction thanks to high accuracy, strong anti-interference ability and high resolution. Target detection technology is one of the key issues in the infrared guidance system, because it can offer the location and shape information of target.This thesis mainly deals with the infrared target detection under complex background, which is achieved by the methods of image enhancement, image segmentation and object recognition. For the poor contrast and low SNR of infrared images, a morphology-based enhancement approach is employed to increase the contrast and improve the SNR. The advantage of the morphology-based enhancement approach is to prevent from blurring the edge of the target, which is conducive to the following image segmentation. For image segmentation, several classical image segmentation algorithms are introduced, and a new multi-feature fusion image segmentation approach is proposed to segment the target region. The proposed approach firstly find the region of interest (ROI) according to the illumination characteristics and geometric properties of the target, then target is segmented in the ROI by combining gray, region and boundary information. The proposed algorithm can detect target from complex background and its performance is superior to the general algorithms. To recognize the target segmented, a new object recognition method based on Hu Moment invariants and Geometric features is proposed. The Hu Moment invariants and geometric features are calculated in binary training object set and binary test set. Firstly, the Hu Moment invariants and geometric features are calculated in binary training object set and binary test set, then standard features and distance threshold are calculated by weighted average method and maximum distance method. Finally, we can recognize the object by distance threshold function. The experiment results show that the proposed methods can obtain the high recognition rate and is easy to implement.