Multiscale Geometric Analysis Theory And Application Research Of Infrared Image

Infrared dim target detection plays an important role in modern infrared detection and remote surveillance system. An in-depth study of these methods has very important significance. In this thesis, we research the infrared small target detection methods based on Multiscale Geometry Analysis(MAG). The main work is through in-depth study on the theory of Multiscale Geometric Analysis and applies it to the field of small target detection in infrared image. The main work of the thesis as follows:(1) Study on the basic principles and characteristics of five classic Multiscale Geometric Analysis methods such as Ridgelet transform, Curvelet transform, Contourlet transform, NSCT and Shearlet transform and lay the theoretical foundations for the follow-up studies. Analysis infrared small target detection technique and simulation several common background suppression algorithms. Draw advantage of infrared weak and small targets detection method based on Multiscale Geometry Analysis.(2) For the problem of small infrared target detection with background interference and noises, a small infrared target detection algorithm based on Nonsubsampled Contourlet transform(NSCT) is put forward. Firstly, the infrared image is preprocessed by conjoint frame subtraction. Then, the NSCT is implemented for the preprocessed image, Using the method combined the improved Nonlinear map function and cross fusion the energy to process the high frequency coefficients so that residual background and noises are eliminated. Finally, threshold segmentation using the Otsu algorithm to isolate small target. The experimental results show that the algorithm can detect small infrared target efficiently, and it can greatly improve the performance of detect dim small target.(3) For the problem of small infrared target detection, a small infrared target detection algorithm based on Shearlet transform is put forward. Firstly, the Shearlet transform is implemented for the original image. Then, According to the uniqueness of dim target in Shearlet modulus and Kurtosis coefficients, the method using fusion multiscale coefficients modulus and the Kurtosis is used for background suppression in order to eliminate background and noises. Finally, threshold segmentation using self-adaptive threshold segmentation technology to isolate small target. The experimental results show that the algorithm can detect small infrared target efficiently.