Systems and methods for segmentation of report corpus using visual signatures are disclosed. According to one embodiment, a computer-implemented method comprises converting a document to a grayscale image and removing noise from the grayscale image by eroding isolated pixels. Connected regions in the grayscale image are determined and a region of the grayscale image having a square shape is identified. An area of the region is computed and if the area is larger than a threshold, determining that the document contains a form.