Siamese Network with Gabor Filter for Recognizing Handwritten Digits

Even though significant progress has been made in the field of pattern recognition in recent years, handwritten digit recognition remains an exciting challenge. Due to its widely possible application, this issue has attracted much attention. Nowadays, different methods are available for handwritten digit recognition. The current demand is for researchers to find new techniques for better performing handwritten digit recognition problems. Therefore, this paper aims to propose a new approach, the Siamese network with Gabor filter, for handwritten digit recognition. Inspired by several studies on the Siamese network and Gabor filter separately, which have already achieved superb performances, this research will bring out the best qualities of their fusion. The computational experiments have been conducted on a handwritten digit image of the MNIST dataset. Empirically, the results implied that the proposed Siamese network with the Gabor filter algorithm outperformed the classical Siamese network and other existing methods in terms of accuracy.