Text-independent writer identification is challenging due to the hugevariation of written contents and the ambiguous written styles of differentwriters. This paper proposes DeepWriter, a deep multi-stream CNN to learn deeppowerful representation for recognizing writers. DeepWriter takes localhandwritten patches as input and is trained with softmax classification loss.The main contributions are: 1) we design and optimize multi-stream structurefor writer identification task; 2) we introduce data augmentation learning toenhance the performance of DeepWriter; 3) we introduce a patch scanningstrategy to handle text image with different lengths. In addition, we find thatdifferent languages such as English and Chinese may share common features forwriter identification, and joint training can yield better performance.Experimental results on IAM and HWDB datasets show that our models achieve highidentification accuracy: 99.01% on 301 writers and 97.03% on 657 writers withone English sentence input, 93.85% on 300 writers with one Chinese characterinput, which outperform previous methods with a large margin. Moreover, ourmodels obtain accuracy of 98.01% on 301 writers with only 4 English alphabetsas input.
展开▼