Traditional pixelwise method (left panel) and our patchwise method (right panel). In the left panel, the point source and the bright region in the background have similar intensities; thus, they have similar saliency levels in the saliency map obtained via a typical existing method. They are similarly judged as point sources or backgrounds in the subsequent pixel-by-pixel thresholding-based segmentation step, resulting in an FP (where the bright region is judged as a point source) or an FN (where the point source is judged as background). The typical point source intensity distribution is an approximate Gaussian distribution. Additionally, the point source is isolated, as it is very different from a bright region that connects with other bright regions. By utilizing this overall information and the contextual semantics of the point source, our proposed method accurately matches and localizes the point source while rejecting the bright region via templates (right panel).