Camera Based Document Image Retrieval with More Time and Memory Efficient LLAH

In this paper, we propose improvements of our camera-based document image retrieval method with Locally Likely Arrangement Hashing (LLAH). While LLAH has high accuracy, efficiency and robustness, it requires a large amount of memory. It is also required to speed up the retrieval of LLAH for applications to real-time docu- ment image retrieval. For these purposes, we introduce the following two improvements. The first one is reduction of the required amount of memory by removing less important features for indexing from the database and simplifying structure of the database. The second improvement is to reduce exploring alternatives during the retrieval process. From the experimental results, we have confirmed that the proposed improvements realize reduction of the required amount of memory by about 80% and that of processing time by about 60%.

[1]  Andrew Zisserman,et al.  Video Google: a text retrieval approach to object matching in videos , 2003, Proceedings Ninth IEEE International Conference on Computer Vision.

[2]  David S. Doermann,et al.  Camera-based analysis of text and documents: a survey , 2005, International Journal of Document Analysis and Recognition (IJDAR).

[3]  Masakazu Iwamura,et al.  Use of Affine Invariants in Locally Likely Arrangement Hashing for Camera-Based Document Image Retrieval , 2006, Document Analysis Systems.

[4]  David S. Doermann,et al.  The Indexing and Retrieval of Document Images: A Survey , 1998, Comput. Vis. Image Underst..

[5]  Haim J. Wolfson,et al.  Geometric hashing: an overview , 1997 .

[6]  Majid Mirmehdi,et al.  Recognising text in real scenes , 2002, International Journal on Document Analysis and Recognition.

[7]  Christoph H. Lampert,et al.  Oblivious Document Capture and Real-Time Retrieval , 2005 .

[8]  Masakazu Iwamura,et al.  Improvement of Retrieval Speed and Required Amount of Memory for Geometric Hashing by Combining Local Invariants , 2007, BMVC.