Semi-Blind Source Scanner Identification

In this paper, we focus on the identification of a scanner at the origin of a given scanned document provided by a public or private organization. Such documents have generally a particular form and content that can be used to simplify the identification process. The unavailability of a precise model of the noise produced by a scanner make it hard to identify it with high accuracy. Previously, source scanner identification schemes proposed to use different well-known filters which are not perfectly adapted to solve this problem. The proposed approach is designed to suppress the maximum of the image's content taking advantage of the original document form. By doing so, the extraction of the scanner noise or signature from acquired images is achieved with better performance than with the filters used in literature. Once the signature extracted, we employ the Euclidean distance to identify the source scanner of a scanned document under test. Experimental results achieved on documents corresponding to registration forms, filled by hand and then scanned with 5 common use scanners illustrate the superior classification performance when compared with the most recent methods. We also demonstrate the robustness of the proposed solution against lossy compression.