Infrared Dim And Small Target Detection Algorithm Under Complicated Background

The infrared detection technology makes full use of the infrared radiation difference between the target and the background for target detection. It is widely used with the advantages of being suitable for all-day work, good concealing, strong anti-interference ability and so on. However, the target imaging area from a distant location is too small, and its strength is relatively weak. The target is almost submerged by background when the background is complicated in infrared image. These all bring a huge difficulty to the target detection. Therefore, the research of infrared dim and small target detection technology under complicated background is very important to both theoretical significance and practical value.Firstly, the infrared radiation characteristics between target and background in infrared image are analyzed. Multiscale geometric analysis is also utilized to interpret the different expression forms of target and background at different scales and different directions. This will lay the theoretical basis for proposing new target detection algorithms for the following.Secondly, the application of partial differential equations in signal and image processing is introduced. The anisotropic diffusion equation has obvious anisotropy, so it is utilized in filtering by discretization. We make it suitable for target detection by modifying the diffusion coefficient, and propose an improved anisotropic diffusion difference filtering by subtracting the filtered image from the original image.Then, nonsubsampled Contourlet transform and singular value decomposition are introduced into the infrared dim and small target detection. A target detection algorithm based on nonsubsampled Contourlet transform and singular value decomposition is proposed. The original infrared image is decomposed by singular value decomposition. The background in the infrared image is predicted by choosing a certain number of the singular value to reconstruct the image, and it is subtracted from the original image to suppress the background. The subtracted image is decomposed by nonsubsampled Contourlet transform, and singular value decomposition is used again to retain the target information, suppress background and filter out noise. The real infrared images are used for the simulation experiment, and the results prove that the proposed algorithm is effective.Finally, Surfacelet transform is demonstrated in detail. Another target detection algorithm is proposed on the basis of Surfacelet transform and anisotropic diffusion equation. The original infrared image is decomposed by the Surfacelet transform into a series of high-frequency subbands and a low-frequency subband. The anisotropic diffusion difference filtering and the local mean removal methods are adopted to refine the high-frequency and the low-frequency subbands respectively. Reconstruct the refined subbands by the inverse Surfacelet transform to obtain the final target detection. The real infrared images are used for the simulation experiment, and the results show that the proposed algorithm can obtain a better performance for the target detection under complicated background robustly.