A Fast System for Dropcap Image Retrieval Mathieu

A Fast System for Dropcap Image Retrieval Mathieu Delalandre and Jean-Marc Ogier L 3 i, La Rochelle University, France mathieu. delalandre@univ-lr. fr Madonne Talk (Tours University) 7 th November 2006

Short CV Madonne Talk (Tours University) 7 th November 2006

Short CV n Personal Information ¡ n Academic Degrees ¡ ¡ n Mathieu Delalandre, 32 years old, Married 1995 -1998 -2001 Lic. Sc In Industrial Computing M. Sc in Computer Science Rouen University, France Research Experiences (5 years, Graphics Recognition) ¡ ¡ ¡ 04/01 -09/01 10/01 -04/05 05/05 -09/05 10/05 -10/06 11/06 -12/06 01/07 -12/09 Madonne Talk (Tours University) 7 th November 2006 Master Ph. D Post-doc PSI Laboratory (Rouen, France) SCSIT (Nottingham, England) L 3 i Laboratory (La Rochelle, France) PSI Laboratory (Rouen, France) CVC (Barcelone, Spain)

Introduction - Old books - Old graphics retrieval - Our problem Madonne Talk (Tours University) 7 th November 2006

- Old books - Old graphics retrieval - Our problem Introduction Old books of XV° and XVI° centuries ¡ Samples ¡ Example of digitized database (BVH, CESR Tours) Book Page ¡ Bartolomeo (1534) Laurens (1621) Old Graphics dropcap figure headline Madonne Talk (Tours University) 7 th November 2006 headline 1385 Graphics Alciati (1511) 46 4755 (3. 4/page) Foreground pixel 63% textual 37% graphical Graphics type 41% dropcap 59% others

- Old books - Old graphics retrieval - Our problem Introduction Old graphics retrieval n System overview ¡ General architecture Image Database ¡ Indexing Retrieval Index Manual Index Extraction Uttama’ 05 Document layout MST Retrieval criterion letter (c) topic (vegetal) Pareti’ 05 Graphics style Zip law Comparison Query ¡ Samples pattern (cross) Baudrier’ 05 Sub image Hausdorff distance Bigun’ 96 Stroke image Radiogram orientation Madonne Talk (Tours University) 7 th November 2006

- Old books - Old graphics retrieval - Our problem Introduction Our problem (1/2) n Context ¡ ¡ n Vascosan 1555 Marnef 1576 MAsse de DOnnées issues de la Numérisation du patrimoi. NE (MADONNE) Project Bibliothèques Virtuelles Humanistes (BVH) du Centre d’Etudes Supérieures de la Renaissance (CESR) Wood Plug Tracking 1555 -1578 Class 1 Wood plug (bottom view) Class 2 printing Class 3 tampon exchange 1511 -1542 copy 1531 -1548 1497 -1507 Madonne Talk (Tours University) 7 th November 2006 Printing house

- Old books - Old graphics retrieval - Our problem Introduction Our problem (2/2) n Problem features ¡ No scaled, no oriented Noise ¡ Offset ¡ n Descriptor choice descriptors Complexity ¡ To image [Gesu’ 99] Ø Template matching, Hausdorff distance Ø no scaled and orientation invariant complex global Ø global (scene) To scalar [Loncaric’ 98] Ø Hough, Radon, Zernike, Hu, Fourrier Ø Scaled and fast local orientation invariant Ø fast Ø local Image Database Formatting ¡ Accuracy Compression ¡ Centering and Comparison Scalability Madonne Talk (Tours University) 7 th November 2006 Query R 1 R 2 R 3

Our system Formatting Compression Madonne Talk (Tours University) 7 th November 2006 Centering and Comparison

Formatting Our system Compression Formatting n Digitalization problems [Lawrence’ 00] ¡ Problem sources n n n ¡ n OLDB (Ornamental Letters Database) ¡ Several image providers Several digitalization tools Length of process Human supervised … query Before (oldb. jpg) Files gray and colour Formats Jpeg and Tiff Compression Packbits and Jpeg Resolutions ¡ 377. 7 Mp Model analysis 2803 Size QUEID « QUery Engine on Image Database » Expertise ? ; from 72 to 450 dpi After Format Madonne Talk (Tours University) 7 th November 2006 279. 7 Mp Model gray Formats charts 2038 Tiff Compression Uncompress Resolutions Diagnostic Base Files Size QUEID Centering and Comparison 250 to 350

Formatting Our system Compression n Run based compression ¡ Run Length Encoding (RLE) n OLDB results n n ¡ image Compression rate RLE Types foreground background Madonne Talk (Tours University) 7 th November 2006 Fixed threshold binarisation Both RLE 0. 88 0. 75 ¡ both Centering and Comparison 0. 95

Formatting Our system Centering and comparison n Centering and Comparison Compression Centering n OLDB results image database query image Size k. pixel Min x 1 Mean 137. 7 337. 06 600. 8 903. 62 Time s Min x 1 176. 67 Size k. run x 1 7. 74 Max Comparison n Time s 1. 1 22. 32 Mean 15. 5 41. 68 Max 87. 8 137. 06 line (y) image 1 line (y+dy) image 2 x pointeur stack x 2 while x 2 x 1 handle image 2 while x 1 x 2 handle image 1 Madonne Talk (Tours University) 7 th November 2006 x 2 x 2

In progress Madonne Talk (Tours University) 7 th November 2006

In progress Our problem n ¡ ¡ n Current time : 40 s Wished time : < 4 s First system Level 1 : image sizes Level 2 : black, white pixels Level 3 : RLE comparison ¡ ¡ ¡ To use a system approach To use a lossless compressio n n n Selection algorithm Key idea Speed query 1 st Level 2 2 sd Level if 1 - 2 < 0 push x, cluster while 1 - 2 < 0 1 Depth Madonne Talk (Tours University) 7 th November 2006 next

In progress n OLDB results n Depth % Min 4% Mean 24% Max 59% n To decrease variability To work on selection To add a level Madonne Talk (Tours University) 7 th November 2006 Run based signature

In progress n Query example n Performance evaluation Query Criterion ? - Scalability - Accuracy - Time processing Benchmar k system Same plug 0. 1947 0. 2517 0. 3485 0. 3616 0. 3819 0. 4064 To produce Next plug 0. 4109 0. 4209 Bench 1 control Bench 2 retrieve IHM Retrieve engine Base display driven labelling Madonne Talk (Tours University) 7 th November 2006 Labels Bench 2

Conclusions and perspectives Madonne Talk (Tours University) 7 th November 2006

Conclusions et perspectives n Conclusions ¡ ¡ ¡ Dropcap image retrieval « wood tracking » Formatting image database (QUEID) Fast approach, two features n n ¡ n RLE comparison ( 7 to 9) Top-down strategy ( 2 to 20) Results 10 s for 2000 images (300 Mo) Perspectives ¡ ¡ Working on RLE signature Benchmark system for performance evaluation Madonne Talk (Tours University) 7 th November 2006

Bibliography Madonne Talk (Tours University) 7 th November 2006

Bibliography 1. 2. 3. 4. 5. 6. J. Bigun, S. Bhattacharjee, and S. Michel. Orientation radiograms for image retrieval: An alternative to segmentation. In International Conference on Pattern Recognition (ICPR), volume 3, pages 346 -350, 1996. V. D. Gesu and V. Starovoitov. Distance based function for image comparison. Pattern Recognition Letters (PRL), 20(2): 207 -214, 1999. S. Loncaric. A survey of shape analysis techniques. Pattern Recognition (PR), 31(8): 983 -1001, 1998. R. Pareti and N. Vincent. Global discrimination of graphics styles. In Workshop on Graphics Recognition (GREC), pages 120 -128, 2005. S. Uttama, M. Hammoud, C. Garrido, P. Franco, and J. Ogier. Ancient graphic documents characterization. In Workshop on Graphics Recognition (GREC), pages 97 -105, 2005. E. Baudrier, G. Millon, F. Nicolier, and S. Ruan. A fast binary-image comparison method with local-dissimilarity quantification. In International Conference on Pattern Recognition (ICPR), volume 3, pages 216 - 219, 2006. Madonne Talk (Tours University) 7 th November 2006

Thanks … Madonne Talk (Tours University) 7 th November 2006