2008-IUPR-07Aug_0818.pdf

0.0 0.1 0.4 100 1000 10033 10622 11.8 1162 1173 12146 12200 123 129 14.2 188 195 199 1993 1998 1999 2.1 2.2 2.3 200 2001 2002 2003 2005 2007 2008 2370 2840 299 2GHz 3.3 3.9 328 370 382 397 401 50.2 51.0 53.1 551 555 583 59.0 596 60.7 61.1 660 788 819 823 962 969 98.4 Abstract Accepted Also Analysis Apr Artificial Aug Automatic BMBF Background Baltimore Bangelore Beijing Beusekom2 Books Both Brazil Breuel Breuel1 Center China Chou Column Computer Conclusion Consider Curitiba D03 DAS DFKI DRR Daniel Decompose Differences Docstrum Document Due EM-like EMlike Each Education Equation Equations Example Examples Experimental Experiments Faisal Federal Field Figure Finally First For Furthermore GUI Gao Gaussian Gaussians Geometric German Germany Given Google Gorman Group Haralick Hence Hindi However ICCV ICDAR IEEE IPeT IUPR Image Inc India Individual Instead Intelligence Interestingly Introduction Iwata Jan Joost Jose Kaiserslautern Kanungo Keysers Keysers1 Kise Last Layout Learning Liang MRF Machine Manhattan Many Mao Markov Matching Ministry Model Modeling Moreover Note Oct One Other Our Overall Parameters Pattern Performance Phillips Princeton Proc Processing Random Recognition References Representation Research Results Robust SPIE San Sato Section Segmentation Sep Shafait Shafait1 Shilman Similarly Since Spitz Starting Stat Statistical Stochastic Structural Style-directed Symp Table Technical Technology The Then These This Thomas Tokuyasu Trans Turbo Two UW-III Unconstrained Understanding University Using VIII Variability Vincent Viola Vision Voronoi Wang X-Y able absolute according accuracy accurate achieve achieved achieving adaptability adaptation adaption added additional advantage alarms algorithm algorithms allowing allows amounts analysis ancestors appear applications applied approach approaches area arg arrows assumed assumption assumption-free assumptions attempts automatic automatically available avoid avoiding background based baseline benchmarking best better blocks book books bound branch brute-force built candidate capable capture case cases center certain challenge chapter check chose class closely coarse column columns combination combinatorial common compared complex complexity component components computed concludes confidence connected considered consisting constitute constraint contains content contrast contribution coordinates copyrights correct correctly correctness correspond corresponding corresponds count counted cover created current cut cuts daniel.keysers data dataset datasets default define defined defining demonstrate demonstrated depend dependency described detected detection develop deviations dfki.de diagram different differing discarded distribution distributions divided dividing document documents does drawback e.g effectiveness element employed error errors estimates evaluation exactly experiment experiments expired explains explicit exponential extending extracted extraction extractor fail failed faisal.shafait false falsely far finding finds fixed focus focused followed following footnotes foreground framework fulfilled funded future gap gaps general generalpurpose generating generation generative generic geometric given gives global globally goal going good grammars grammatical ground ground-truth grouping half hand happen height help hierarchical hierarchy high highest highly human i.e illustrations image images implicit important improvement include incorporate independent index indices information initial inspected integer inter interactive interested intersecting involved iteratively iupr.net j-1 joost just known labeled language large layout layouts lead learn learning learns left likelihood likely limit limitation limits line linear lines literature log log-likelihood long low lower main mainly major making manually massive match matched matches matching max maximum mean meaningful measure measures merged met method methods middle minimize missed mixture model model-based modeled modeling models motivated n-tuple need needed new noise non-Manhattan non-generative non-stereotypical normal normalized normalizing novel number observing obtain obtained occur old ones operators optimal optimization order outliers output over-segmenting oversegmentation oversegmentations page pages paper parameter parameters paramter parsing partially particular parts per-cut percentage performance picked plausibility plays point position possible practical practically practice preliminary present presented prevented previous probabilistic probability problem problems process produces programming project propose proposed provide publication purpose quality quantity quite rare readily recent recognition recomputed rectangle rectangles regarded relative released rely reported represent representation representative represented represents require researchers resembles result resulting results return right robust role rule-based running scale scanned scans score scores search second segment segmentation segmentations segmented segments selects separated separator separators sequence set showed shown simply size small smaller solution spacing specific spectrum split square standard stands state-of-the-art statistical statistically step steps stochastic stretch structural structure structures style style-directed submodels subset suitable symbols systems t-1 table target task techniques tends term terminal terminates terms tested tests text text-line time title tmb total train trainable trained training tree tries trim truth trying tuning type types unconstrained underlying undersegmentation undersegmentations undersegmented understanding updated updating usability use used usefulness user uses using usually valid value values van variability variance variances variations vary vertical visualization visualized volume want white whitespace whitespaces widely width word work work1 written wrong wrongly x-center y-center zone zone-level