Research On Infrared Image Enhancement, Segmentation And Target Tracking

Infrared imaging guidance has been an important development direction in the field of precision guidance. Studying infrared image enhancement, segmentation and target tracking techniques is important for improving performance of infrared imaging guiding system. On the basis of previous research results, research on infrared image enhancement, segmentation and target tracking techniques has been done in this paper, and described as follows:Firstly, a method based on stationary wavelet transform and Retinex is discussed. The low frequency subband image of the largest scale is enhanced by multiscale Retinex algorithm, and the gain coefficients of high frequency subbands are available by calculating local contrast of the enhanced low frequency subband based on fuzzy rules, thus getting enhanced high frequency subband images. Experimental results show that, the whole visual effect is improved significantly by the method, and enhancement effect is more obviously for noisy images with non-uniform brightness. Then an infrared image enhancement method based on Nonsubsampled ContourletTransform(NSCT) and fuzzy logic is given. The original infrared image is decomposed by NSCT. The lowpass subband coefficients are enhanced by fuzzy enhancement algorithm and the bandpass subband coefficients are enhanced by nonlinear enhancement algorithm. Experimental results show that the method improves definition, details and visual effect of the image significantly.And then, an infrared image segmentation method based on the within-class absolute difference and chaotic particle swarm optimization is studied.The less within-class absolute difference can make the cohesion performance better, and the area difference between background and target is used to inhibit the tendency of an equal division. Therefore, a more reasonable threshold selection rule is formed comprehensively. As the computational burden of finding optimal threshold vector is large for the two-dimensional thresholding, thus niche chaotic mutation particle swarm optimization is used to find the optimal threshold vector. Experimental results show that the proposed method is effective for smaller target infrared image thresholding and the running time is significantly reduced.Next, an infrared target tracking method based on SIFT and RANSAC is proposed. SIFT is used to extract features of targets, and Euclidean distance is adopted to measure similarity of the feature points , error matches of which are eliminated by RANSAC algorithm. Thus, the accurate target location is calculated by affine transformation model. Experimental results show that the proposed method is superior to the infrared target tracking method based on mutual information measure both in accuracy and real-time performance.Finally, an infrared target tracking method based on mean shift and particle filter is introduced. The mean shift algorithm and particle filter algorithm are combined effectively by adopting mean shift algorithm to perform iterative convergence of particles, thus implementing robust tracking by using a few sampling particles. Experimental results show that the method is more accurate than mean shift tracking method, and can track the target stably under short-term occlusion circumstances.