Hand-Written Characters Recognition using Siamese Network Design

Hand-written documents are being digitalized and various traditional approaches have been merged to automatically recognize the syntax and semantic of target statement. However, real time applications such as bank data processing and security are still challenging due to the variation in human hand-written document. Similarly, traditional assistive scheme requires large amount of data to learn new features and their pattern. To deal with this, one-shot learning scheme is employed to recognize hand-written characters using Siamese network architecture. The Siamese network can compute the similarity or difference between two data samples and also effective to process unseen data that has not been previously trained. Additionally, convolutional neural network architecture is fused up to extract significant features during training and provides the stability in learning. The performance is validated through standard evaluation measures operated on the public dataset.