Reconstructing shredded documents through feature matching Authors: Edson Justino, Luiz S. Oliveira, Cinthia Freitas Source: Forensic Science International 160 (2006), pp. 140–147 Date: Speaker: Meng-Jing Tsai

Different Kinds of Shredding 2

Outline ҉ Introductions ҉ Proposed Method ҉ Experimental Result ҉ Conclusions 3

Introductions The amount of time necessary to reconstruct a document depends on the size and the number of fragments, and it can be measured in days or even weeks. Traditional puzzle solving algorithms usually take into account smooth edges and well defined corners. The act of shredding a piece of paper by hand often produces some irregularities in the boundaries. 4

Proposed Method The block diagram of the proposed methodology

Proposed Method Pre-processing – In order to overcome this kind of problem, we have tested different algorithms, and the one that brought the best results was the well-known Douglas–Peucker (DP) algorithm. 6

Proposed Method Douglas-Peucker Algorithm 7

Proposed Method Pre-processing 8

Proposed Method Feature extraction Fig. 1 Angle features extracted from the polygon (180,110) (180,0) (10,110) (10,70) (55,67) (67,25) ° 9

Proposed Method Matching – Computing the similarity between polygons – Global search 10

Proposed Method Computing the similarity between polygons – If the complementarity is verified like in Fig. 2, then W angles =1. Fig. 2 Similarity between angles 11

Proposed Method Computing the similarity between polygons Fig. 3 Distance features extracted from the polygon 12

Proposed Method We consider the relevance of the matching regarding the perimeter of the fragment using the following rules: – If the contour matched represents more than 1/5 of the perimeter of the fragment, then W matching = W matching +2. – If the contour matched represents more than 1/10 of the perimeter of the fragment, then W matching = W matching +1. – Otherwise, W matching is not increased. 13

Proposed Method Global search – Let us consider a shredded document D = {F 1, F 2,..., F n } composed of n fragments. – The algorithm compares the fragment F 1 with all the other fragments searching for the best matching. Fig. 4 Best matching (a) fragments i and j and (b) new fragment F ij where three vertices were removed 14

Proposed Method Steps of the document reconstruction 15

Experimental Result Examples of a document totally reconstructed: (a) fragments and (b) document reconstructed. 16

Experimental Result Performance of the proposed methodology in reconstructing documents shredded by hand. The fragments size range from 1cm × 1cm to 5cm × 5cm. 17

Conclusions This paper proposed a method for document reconstruction based on feature matching. It can be addressed by choosing the most important aspects for the application. 18

Thank you for your listening. 19