Research On Methods Of Infrared Image Processing And Dim Target Detection

Infrared imaging technology has been widely used in precision guided weapon system due to itsspecific advantages. It is important to study infrared image processing and dim target detectiontechnology for improving the performance of infrared imaging guidance system. Infrared imageenhancement, segmentation, fusion and dim target detection is studied in this paper. The main work isas follows:Firstly, an infrared image enhancement method based on contourlet transform and chaoticparticle swarm optimization (PSO) is studied. The chaotic PSO algorithm is used to search theoptimal parameters of functions for low and high frequency sub-band enhancement. Experimentalresults show that the proposed method can effectively enhance image details and suppress noise, andthe whole visual effect is improved significantly.Secondly, two two-dimensional minimum error image thresholding methods based on chaoticPSO or decomposition are proposed. A chaotic PSO algorithm is used to find the optimal threshold oftwo-dimensional minimum error method. As a result, the redundancy computation is reduced greatly.Then, the computation of two-dimensional minimum error method is converted into twoone-dimensional spaces. The experimental results show that, the two proposed methods can greatlyimprove the running speed while the segmented result is as good as or better than the existingtwo-dimensional minimum error thresholding method.Then, a fusion method of infrared and visual image based on cycle spinning and contourlettransform is implemented. When the complex contourlet transform is used to fuse images, cyclespinning is introduced to overcome the pseudo Gibbs caused by image transform. Compared with themethods based on wavelet or contourlet transform, the information amount of the obtained image ismore plentiful.Next, an infrared dim target detection method based on dual-tree complex wavelet transform(DT-CWT) and kernel principal component analysis (KPCA) is given. The background image isseparated form original image by KPCA. Then the background image is subtracted from the originalimage, the obtained residual image is denoised by DT-CWT. The dim target is segmented from thedenoised residual image by the proposed Tsallis cross entropy method. The method can suppress thebackground and noise better, and get higher detection probability.Finally, an infrared dim target detection method based on neighborhood gray entropy andclassification is explored. The infrared image of dim target is first denoised by complex contourlet.Then, a sliding window is designed to scan the whole image. The pixels in the sliding window aredivided into two categories as object and background. The difference of grayscale means of twocategories and the neighborhood gray entropy of central pixel in the window are used to suppress thebackground. At last, the exponential cross entropy thresholding method is used to detect the dim target.Experiments show that the method is simple, effective and easy to implement by hardware.