Fast detection of small infrared objects in maritime scenes using local minimum patterns

This paper describes a novel approach for fast detecting small maritime objects in infrared (IR) images. It is based on the local minimum patterns (LMP), which are theoretically the approximations of some stationary wavelet transforms (SWT). Using LMP to estimate the background with a single image, we obtain an object-aware saliency map by background subtraction. Regions of potential objects are then segmented by an adaptive threshold based on the histogram of the saliency map. We finally propose a fast clustering algorithm for localizing objects from segmented regions. Extensive experiments on challenging data sets show a competitive performance.