Statistical feature extraction/selection for small infrared target

Abstract

Feature extraction and selection have become necessary steps for `low loss dimension reduction'. Machine learning, data mining and pattern recognition are the respective fields to use this methodology. In small target infrared image many false alarms may occur due to different clutters. In machine learning, for preprocessing set of relevant features of the target is required and for dimensionality reduction selection of most appropriate feature subset is done for the classification purpose. To reduce false alarm rate, this paper focuses on extraction of relevant features of small infrared target where each feature is analyzed statistically and selection of relevant feature subset is done by using forward feature selection approach and there is reduction in false alarm rate by a factor of 2.3 in compare to filter based detection by using classifiers.