| Detection of targets embedded in correlated infrared clutter is considered one of the challenging problems in the automatic target detection and recognition (ATD/R) systems. The problem grows more complex when the targets are small, faint, and obscured by surrounding objects. A number of investigators have shown that using the gray-scale feature only is not adequate for the detection, especially in the above complex situations. Multidimensional feature space has been suggested to be more useful than the gray-scale feature, where a variety of features may be defined over a neighbourhood set around each pixel, e.g., mean, median, variance, commonality, and total variation. Following this suggestion, a need for a good feature extractor is required. The main objective of this feature extractor is to emphasize the important features (targets) in the image, and to deemphasize those which are irrelevant (clutter).; Time-frequency (TF) image representations including non-adaptive transformation (such as short-time Fourier transform (STFT), wavelet transform (WT), and wavelet packet (WP)), and adaptive approaches (such as matching pursuit (MP)) have been used for feature extraction. In this thesis, we use the fast orthogonal search (FOS) algorithm to establish a new detector to detect small targets embedded in infrared clutter. FOS is an adaptive signal representation approach which provides an efficient procedure to represent the signals with any arbitrary functions as well as rejection of white Gaussian noise using an adaptive threshold.; In the thesis, some recent detection algorithms used for small infrared target detection in the ATD/R systems are investigated. These algorithms include the constant false alarm (CFAR) detectors, double-window filter (DWF), and some of the wavelet-based and MP-based detectors.; The main objective of the thesis is to extend the FOS to two-dimension space (2D-FOS) and to propose a 2D-FOS based detector. This new detector has three components: a local whitening filter, the 2D-FOS modelling algorithm, and then the first-order statistical analyses of the potential regions detected by the 2D-FOS are used for further reduction of false alarms.; Experimental results using a database of real infrared images demonstrated that the new detector yields promising results to solve some of the detection problems mentioned above. A comparison between the output performance of the new detector and some of recent detectors (DWF and some of the wavelet-based detection algorithms) are also included. |
