Morph-Cnn: A Morphological Convolutional Neural Network For Image Classification

Deep neural networks, an emergent type of feed forward networks, have gained a lot of interest especially for computer vision problems such as analyzing and understanding digital images. In this paper, a new deep learning architecture is proposed for image analysis and recognition. Two key ingredients are involved in our architecture. First, we used the convolutional neural network, as it is well adapted for image processing since it is the most used form of stored documents. Second, a morphological feature extraction is integrated mainly thanks to its positive impact on enhancing image quality. We have validated our Morph-CNN on multi digits recognition. A study of the impact of morphological operators on the performance measure was conducted.