Seite - 25 - in Document Image Processing

J. Imaging 2018,4, 27 Table2.Binarizationresults for images fromtheLiveMemoryproject. AlgName P(f/f) P(b/b) Time(s) Bilateral 100.00 98.90±1.07 3.3325 IsoData-ORIG 99.56±0.69 98.61±1.99 0.0734 Otsu 99.60±0.68 98.57±2.08 0.0735 Moments 99.99±0.03 97.91±1.87 0.0716 Ergina-Local 98.98±2.82 97.62±1.04 0.9917 Huang 99.93±0.27 96.42±4.20 0.0865 Triangle 100.00 94.24±2.15 0.0728 Mean 100.00 83.58±5.59 0.0747 Niblack 99.76±0.76 78.31±2.97 0.6710 Pun 100.00 55.28±3.60 0.0800 Percentile 100.00 53.91±1.96 0.0795 Kapur-Sahoo-Wong 98.62±4.92 97.15±1.44 0.0729 Figure6. ImagesfromLiveMemorywiththebest ((left)—P(f/f)=100.00,P(b/b)=99.99)andtheworst ((right)—P(f/f)=100.00,P(b/b)=95.97)binarizationresults for thebilateralfilterwithzoomsfromthe original (top) andbinary(bottom)parts. 3.3. TheDIBCODataset This dataset has all the 86 images from theDigital Image BinarizationContest from2009 to 2016. Table3presents theresultsobtained. Theperformanceof thebilateralfilter in thissetmaybe consideredgood, ingeneral. Theoverallperformanceof thebilateralfilterwasstronglydegradedby thesingle imageshowninFigure7 (right) inwhich theP(f/f)of25.93drasticallydroppedtheaverage resultof thealgorithminthis testset. It is important toremarkthatsuchanimageisalmostunreadable evenforhumansandthat itdegradedtheperformanceofall thebestalgorithms. Table3.Binarizationresults for images fromDocument ImageBinarizationCompetition(DIBCO). AlgName P(f/f) P(b/b) Time(s) Ergina-local 91.37±6.25 99.88±1.89 0.1844 RenyEntropy 90.13±14.19 96.77±3.50 0.0125 Yean-Chang-Chang 90.61±14.44 96.16±4.35 0.0112 Moments 90.75±9.91 95.80±5.19 0.0112 Bilateral 92.99±9.06 90.78±16.01 0.6099 Huang 95.62±6.37 84.22±18.36 0.0147 Triangle 96.40±5.72 80.80±23.32 0.0113 Mean 99.35±1.14 78.99±9.35 0.0115 MinError 92.79±23.46 74.29±19.36 0.0115 Pun 99.68±0.82 56.20±6.18 0.0122 Percentile 99.71±0.72 55.06±3.58 0.0121 Sauvola 59.75±30.06 99.58±079 0.6933 Niblack 95.91±2.31 78.61±5.69 0.1241 25