Degradation Classification on Ancient Document Image Based on Deep Neural Networks

摘要: In this paper, we study degradation classification on ancient document images using three pre-trained models of benchmarking CNN architecture, i.e., Resnet101, Mobilenet V2, and Shufflenet. We use Document Image Binarization Contest (DIBCO), Persian Heritage Dataset (PHIBD), private Jawi datasets for experimental purposes. grouped the into four categories, namely: bleed-through/showthrough/ink-bleed, faint-text low contrast, smear-spot-stain, uniform degradation. training progress, set optimizer to ADAM, initial learn rate $10 ^{-4}$, epoch values: 5, 25, 50 epoch. To test model, conduct two testing stages: (1) unblind testing, (2) blind testing. The result shows that Shufflenet with 25 achieved 100% 85% accuracy respectively, obtained fastest computational process. concluded could be chosen in classifying degradations based its time.