Detection of Barely Discernible Subpixel Objects Using a Modified Orthogonal Subspace Projection Method

This paper deals with the problem of detection of the video/infrared small barely discernible targets in the presence of the heavy background clutter and channel noise. The majority of well-known detection algorithms are based on the assumption that the signal energy reflected from the targets' surface and acquired by the sensor is more or equal to the background clutter energy. This assumption limits the implementation of these algorithms in the case of low signal-to-clutter ratio. The presence of the subpixel target decreases the background power that hence may be different under the null and alternative hypotheses. To detect the barely discernible targets we derive the Generalized Likelihood Ratio Test (GLRT) under hypothesis dependent structured background power and contrast it with well-known Orthogonal Subspace Projection (OSP) algorithm. Experiments on simulated data and real video images demonstrate the ability of this proposed detector to detect the subpixel barely discernible targets in the presence of the heavy background clutter.