Deep Morphological Gradient for Recognition of Handwritten Digits
Mouhssine El Atillah, Khalid El Fazazy · 2019
The optical character recognition (OCR) is a technology that has evolved enormously in recent years. It is able to read characters with extreme precision. The recognition of handwritten numbers by the algorithms of deep learning is one of those places of research that has been remarkably popular. Recently, some studies are moving towards this side and give remarkable results for such recognition. We shed light on a deep morphological gradient for the problem of recognition of handwritten digits. We use a Multilayer Perceptron Network (MLP) preceded by the morphological gradient algorithm to detect the contours of the digits. This model is applied to the database MNIST of handwritten digits available on Kaggle, which consists of 70,000 images. The classification accuracy of the model was 100% with a very minimum loss of 10-5%.