IR target detection algorithm based on mixture probabilistic kernel principal component jointed with quadratic correlation filter

Based on the feature extraction of principal component, a novel infrared target detection algorithm was proposed which using subspace quadratic synthetic discriminant function (SSQSDF). Firstly, the kernel principal component analysis was extended to mixture probabilistic model, and the latter get the principal component vectors of target samples. Then, training samples and samples to be detected were projected on principal component vectors obtained previously to acquire their low-dimension feature components, and the obtained components are used as the sample parameters for the SSQSDF. The detected samples which had a higher SSQSDF filtering output than given threshold were considered as the detected targets. The proposed algorithm can evidently restrain clutter noise, improve target detection precision. Experimental results under complex scenery demonstrate that the proposed algorithm is feasibility and effectiveness.