to develop image processing techniques for computer-assisted forensic toolmarks identification.
Toolmark identification is an important part of criminal investigation. As criminals become more mobile, law enforcement services need to handle a growing bulk of toolmarks, originating from crime scenes and from tools apprehended in the possession of suspects. Ideally, we would like to match a given toolmark against all known toolmarks in the same category (e.g. crowbar traces) for possible "hits". Such an exhaustive search can only be feasible using automated comparison criteria.
On the other hand, only a human expert can give sworn testimony in a court of law and take responsibility for its accuracy. Consequently, the role of the automated techniques is to screen the accumulated corpus of material evidence for possible matches and supply the human expert with a much smaller set of toolmark pairs for verification. Computers can also facilitate the handling of images on the expert's workbench. They are not expected to replace the expert witness.
The project focused on two types of toolmarks, corresponding to the compression and sliding categories as described by the Isaspro introductory page.
Compression toolmarks (made by pressing or hitting, such as the trace of a seal in wax) are difficult to represent by standardized attributes. The project therefore limited itself to developing a virtual comparison microscope (the SOMIC, as described below).
Sliding toolmarks (also known as striated toolmarks, "stria" meaning a narrow stripe) are when a tool slides over a surface or creates a surface by cutting through a material. A good model is slicing butter with a toothed knife. We chose to regard these marks as "smudged" two-dimensional extensions of one-dimensional grey-level profiles, which in turn represent the physical profile of the tool edge. Consequently, tools were developed to project the image back into a 1D domain and to compare such 1D profiles.
The Vidocq toolmark database stores the toolmarks and submits them to the Correlator. Click here for a description.
SOMIC - the Software Comparison Microscope is an interactive program under Windows which allows two bitmap images to be placed side by side and patchworked together along an arbitrary broken line (shown in red in the picture below).
The Semiautomatic Stria Correlator takes the projected grey level profiles of two striated toolmark images and slides one of them along the other to find the position where their striae coincide most closely. This position, and a measure of the correlation obtained, can then be used to provide the human expert with a short list of good hits.
While it is relatively straightforward to compare the geometric layout of stripes in two images, the individual stripes may or may not actually resemble their supposed analogues. At present, the Correlator relies on just geometric matching, regardless of the form of individual stripes. Optionally, it can use grey level correlation of entire 1D patterns, in which case stripe positions are not explicitly detected. We believe that best results can be obtained by combining geometric criteria with a measure of individual stripe similarity. This measure, however, will have to be more sophisticated than simple grey level correlation, because striae left by the same nick or notch of the tool can have very different grey level profiles.
The Toolmark Expert Interviewer was developed to obtain a corpus of expert judgments on local stria similarities. It shows the expert two small fragments of striated images (so small as to contain, ideally, just one isolated stripe each). The expert is then asked to decide if these two small fragments match. See the picture below for an example of user interface.
The data obtained by the Interviewer are needed to develop specialized stripe similarity measures. One possible way of using them, which we are developing, is by training twin neural networks.
Marple - a public database. As a contribution to the toolmarks processing community, the set of samples accumulated via the Interviewer is available as the MARPLE database.
IITiS - Institute for Theoretical and Applied Computer Science of the Polish Academy of Sciences. Baltycka 5, 44-100 Gliwice, Poland.
Forensic Science Institute of the Dutch Ministry of Justice (former Gerechtelijk Laboratorium van het Ministerie van Justitie) in Rijswijk, The Netherlands
KWP (Regional Police Command) in Bydgoszcz, Poland
Laboratoire I3D (Interaction, Image et Ingénierie de la Décision) of the Université des Sciences et Technologies de Lille, France.
The project is financed as Workpackage # 14 of the European CRIT-2 programme (Cooperative Research in Information Technology).
The International Association for Identification has a number of pages on
forensic identification of shoeprints, fingerprints etc.
Zeno's Forensic Site also has
links to image databases