Reweighted Infrared Patch-Tensor Model With Both Non-Local and Local Priors for Single-Frame Small Target Detection

摘要

Many state-of-the-art methods have been proposed for infrared small targetdetection. They work well on the images with homogeneous backgrounds andhigh-contrast targets. However, when facing highly heterogeneous backgrounds,they would not perform very well, mainly due to: 1) the existence of strongedges and other interfering components, 2) not utilizing the priors fully.Inspired by this, we propose a novel method to exploit both local and non-localpriors simultaneously. Firstly, we employ a new infrared patch-tensor (IPT)model to represent the image and preserve its spatial correlations. Exploitingthe target sparse prior and background non-local self-correlation prior, thetarget-background separation is modeled as a robust low-rank tensor recoveryproblem. Moreover, with the help of the structure tensor and reweighted idea,we design an entry-wise local-structure-adaptive and sparsity enhancing weightto replace the globally constant weighting parameter. The decomposition couldbe achieved via the element-wise reweighted higher-order robust principalcomponent analysis with an additional convergence condition according to thepractical situation of target detection. Extensive experiments demonstrate thatour model outperforms the other state-of-the-arts, in particular for the imageswith very dim targets and heavy clutters.