Image Averaging and Shape Recognition in Forensic Analysis of Video Evidence
Category: Scientific Publication
Publication: Int. J. Forensic Engineering, 2023
Authors: Geoffrey T. Desmoulin, Marc-André Nolette, Theodore E. Milner
Summary: Averaging of raw images obtained from single frames of a body-worn camera was used to illustrate the ability of averaging to significantly reduce the noise present in a single video frame. The outline of the subject in the averaged image was then compared to outlines of a different subject adopting four similar postures, which included one posture that mimicked the posture of the first subject (reference posture). Using Hausdorff distance as a measure of shape similarity, it was possible to demonstrate that the outline of the second subject was most similar to the outline of the reference posture when the second subject mimicked the reference posture. Despite the similarity of the four postures, the Hausdorff distance was at least 50% less when the second subject mimicked the reference posture than for any of the other postures.
