Improved segmentation for footprint recognition of small mammals

Abstract

In this paper we improve the automatic extraction of segments by resolving some of the issues for collected rat footprints, such as incomplete, fading, merged, or overlapping prints, or cuts due to the applied rectangular clipping process. First, binarization is by an adaptive method (proposed by Otsu) on the given input segment. Second, we remove small artefacts with a subsequent adaptive method. Third, merged regions are separated by a morphological method using an adaptive mask. Next, we find meaningful pads (central pad or toes) by analysing geometric relations defined by triangulation. Finally we reconstruct damaged footprints by using a convex-hull algorithm. We present experimental results of reconstructed footprints, and distributions of extracted features for improved segments. In the proposed technique, we automatically improve the quality and reliability of a scanned footprint image so as not to lose potential information for subsequent identification steps.