Selection and extraction of difference statistical features based on dual color MWIR images

Feature space dimension of the dual-color mid-wave infrared (MWIR) image is large and difficult to conduct a comprehensive analysis of its dualband imaging differences. The three-level feature selection model based on the statistical features was proposed. This model firstly standardized the dual-bands original statistical features, then with the distribution of four-group dual-band images, through the analyses of the significant-difference level, consistentdifference level and special-difference level, chose the effective statistical features which satisfied the dual-band image differences distribution regularity. Based on the feature selection, this paper used locally linear embedding (LLE) to reduce the dimension of the effective statistical features. The experimental data indicate that the one-dimensional features after feature selecting and extracting can show the differences in rate of dual-band image, achieving the purpose of using small amount features to comprehensively analyze the differences in dual-band.