Research On Methods For Feature Extraction Of Infrared Image Targets

Feature extraction of infrared images targets are one of the key technologies ininfrared early warning and imaging guidance. Infrared systems have good concealment,strong anti-interference ability, far operating range and fast search speed. Besides, theyhave no mirror echo and can work day and night. All of these advantages make themused in military or civilian fields more and more. Consequently, how to extract thedistinct features, to improve the detection and recognition of infrared images and thus toimprove the intelligent level of the overall infrared systems, has been the researchdirection and focal point for the domestic and foreign scholars.This paper focuses on the methods of image feature extraction and getting higherrecognition rate for the classification tasks. Firstly, we briefly introduce theories ofprinciple component analysis (PCA), independent component analysis (ICA) andindependent subspace analysis (ISA), and study their characteristics that be representedon the images. Further, by constructing the correspondent image data, different forms ofexpression of these three methods are directly come out. Secondly, for the randomnessof the extracted independent components, sparse ranking algorithm of these independentcomponents is presented. The experimental results show that less advantageouscomponents can be used to obtain better recognition accuracy and then the real-timerequirements for the real system can be met. Thirdly, the statistical characteristic of theISA for natural images are analyzed in detail. What’s more, we apply it to the featureextraction of the infrared images and the recognition rate is satisfying. Finally, using theclassifier based on distance measurement, the global and local features of ICA areutilized for classification respectively, and then the larger dispersion is taken as the rightone. Experimental results demonstrate the validity of the proposed fusion recognitionmethods for the infrared image recognition.