Offline signature identification using the histogram of symbolic representation
Mohamed Anis Djoudjai, Youcef Chibani, Nassim Abbas · 2017
This work proposes an off-line handwritten signature identification system using the Histogram of Symbolic Representation (HSR). The HSR is considered as one-class classifier which has the ability to generate a model for each writer using only its own reference signatures. This method allows also modeling the writing style of each writer by taking into account the variability of signatures. To evaluate the robustness of the proposed identification system, two well-known standard offline handwritten signature datasets are used, namely CEDAR-55 and GPDS-300. The experimental results achieve an accuracy of 98.63% and 97.84%, respectively. These obtained results are broadly better than the state-of-art when using only 5 reference signatures.