Infrared Small Target Detection Based On Multiresolution Analysis And Independent Component Analysis

The small target detection in infrared image is a crucial technique for the precision guided weapon. It is of much theoretical and practical significance to achieve fast and accurate detection of the small target under complex background. On the basis of introducing domestic and foreign development of small target detection, this paper studies the detection methods of small infrared target based on multiresolution analysis and independent component analysis. The main tasks are as follows:Firstly, the detection methods of infrared small target based on the lifting wavelet transform/dual-tree complex wavelet transform and morphology are proposed. The wavelet transform is replaced by the lifting wavelet transform / dual-tree complex wavelet transform, and Top-hat operator is used in these methods, so that the background of infrared small target image is suppressed and the contrast between target and background is enhanced. The experimental results show that these methods can detect infrared small target accurately, and detection results are superior to those of the detection method based on wavelet transform and morphology.Secondly, aimed at the detection problem of dim target in infrared image that contains background interference and noise, a detection method is proposed based on dual-tree complex wavelet transform and independent component analysis. The background image separated by independent component analysis is subtracted from the original image first, then the fast detection of infrared small target is realized combining dual-tree complex wavelet transform. This method has higher detection probability, and is better than the detection method based on fast independent component analysis.Thirdly, the detection method of infrared small target is realized based on nonsubsampled contourlet transform and independent component analysis. First, original image is reprocessed by nonsubsampled contourlet transform and independent component analysis. Then, the preprocessed image is segmented by the threshold selection algorithm based on the within-class variance and area difference between background and target. The experimental results show that this method has better detection performance than the detection method based on NSCT.Fourthly, a detection method is introduced based on complex contourlet transform and principal component analysis. The background image separated by principal component analysis is subtracted from the original image. The residual image is denoised by complex contourlet transform and filtered by Top-hat operator. Then the preprocessed image is segmented by using the threshold selection algorithm based on fuzzy Renyi entropy, and the fast detection of infrared small target is achieved.Fifthly, a detection method is proposed based on non-negative matrix factorization and independent component analysis. The background of original image is suppressed by non-negative matrix factorization and independent component analysis, respectively, and different residual images of small target are obtained. The residual images are denoised by using complex contourlet transform. The preprocessed image is obtained by the addition of the above-mentioned denoised residual images. The thresholding methd based on fuzzy gray entropy is used to segment the preprocessed image for the detection of dim target under complex background. The experimental results show that this method can effectively detect the target in an infrared image with complex background.Finally, thresholding methods for the segmentation of infrared target image are proposed based on the symmetric cross-entropy and the area difference between background and target. First, 1-D threshold selection formula is derived in this paper. Then it is extended to 2-D. The thresholding method and the simplified thresholding method based on oblique segmentation of 2-D histogram are given, which can improve the anti-noise performance greatly. The experimental results show that, proposed methods are more effective in segmenting infrared small target image, compared with the thresholding methods of the maximum entropy, the Otsu and the asymmetric cross-entropy based on oblique segmentation of 2-D histogram.