Deep Learning for Philately Understanding

Deep Learning has enabled incredible advances in the field of computer vision in the recent years. Convolutional neural networks have demonstrated the ability to classify images into thousands of classes with high precision. These neural networks have solved various computer vision related problems like object classification, object detection, and image segmentation. Deep learning applications have benefited fields ranging from security to cancer treatments, and in this paper we apply deep learning to the niche field of philately. The objective of this research is to recognize stamp features such as plate number and row and column letter using photographs of stamps. We use the Queen Victoria Penny Red stamp Scott 33, a legacy stamp issued in mid-nineteenth century Britain. This stamp is a hugely popular stamp among philatelists. For this research, approximately 130,000 of these stamps have been meticulously photographed and cataloged. Despite sometimes poor quality and strong cancellation marks, results demonstrate over 95% accuracy on row/column recognition and 90% plate recognition.