Signature-based Document Image Retrieval

As the most pervasive method of individual identification 4 4 and document authentication, signatures present convincing evidence 5 5 and provide an important form of indexing for effective document image 6 6 search and retrieval in a wide range of applications. In this work, we 7 7 developed a fully automatic signature-based document image retrieval 8 8 system that handles: 1) Automatic detection and segmentation of signa9 9 tures from document images in unconstrained setting and 2) Translation, 10 10 scale, and rotation invariant signature matching for document image re11 11 trieval. We treat signature retrieval in the broader context of non-rigid 12 12 shape matching and retrieval, and quantitatively study shape represen13 13 tations, shape matching algorithms, measures of dissimilarity, and the 14 14 use of multiple query instances in document image retrieval. Extensive 15 15 results using large real world collections of English and Arabic machine 16 16 printed and handwritten documents demonstrate the excellent perfor17 17 mance of our system. To the best of our knowledge, this is the first auto18 18 matic retrieval system for general document images by using signatures 19 19 as queries, without manual annotation of the image collection. 20 20