Seite - 96 - in Document Image Processing

J. Imaging 2018,4, 41 NetworkArchitectures LayerType LayerSize TrainableParameters TotalParameters NA-4 Conv1layer 64×64×64 1088 14,514,735 Conv2 layer 64×33×33 65,600 Dense layer 1000 14,401,000 Output layer 47 47,047 NA-5 Conv1layer 32×64×64 544 1,649,423 Conv2 layer 32×33×33 16,416 Conv3 layer 32×17×17 16,416 Dense layer 1000 1,569,000 Output layer 47 47,047 NA-6 Conv1layer 64×64×64 1088 3,316,335 Conv2 layer 64×33×33 65,600 Conv3 layer 64×17×17 65,600 Dense layer 1000 3,137,000 Output layer 47 47,047 Table 4. In this table,we report the results in termofmaximum,minimum,mean, and standard deviation recognitionaccuracyobtainedwithdifferentnetworkarchitectureson ISIDCHARwhen thesystemtrainedfor50epochswith theRMSPropoptimizer. Thebest scoresare inbold. RecognitionAccuracy DifferentNetworkArchitectures NA-1 NA-2 NA-3 NA-4 NA-5 NA-6 Maximum 0.8572 0.8641 0.903 0.9079 0.9311 0.9602 Minimum 0.7093 0.7475 0.7711 0.7788 0.7422 0.8067 Average 0.8383 0.8501 0.8927 0.8941 0.9150 0.9463 Std.Deviation 0.0321 0.0232 0.0210 0.0197 0.0308 0.0252 (SRFK 1$ 1$ 1$ 1$ 1$ 1$ Figure4. Inthisfigure,wedrawtherecognitionaccuracyobtainedwithdifferentnetworkarchitectures onthe ISIDCHARdatabaseateachepoch. TheRMSPropoptimizerwasused. The best recognition accuracy of the ISIDCHARdatabasewas obtainedwithNA-6 network architecturewithRMSPropoptimizer.However, itmaybepossible that thisnetworkcouldperform betterwithother optimizers. To further investigate,weperformedexperimentswith sixdifferent optimizers. Table 5 shows the recognition accuracy obtainedwithNA-6 at different optimizers. Thehighest recognitionaccuracy96.02%wasrecordedwithNA-6atRMSPropoptimizer. TheAdam 96