Deep CNN Framework for Classification and Feature Extraction

Abstract

The most important task of computer vision is not limited to classification, object detection, alignment, and tracking. Recently, convolutional neural networks (CNNs) have emerged as the most prominent techniques to achieve computer vision tasks. This chapter presents theoretical and practical aspects of CNNs using deep learning. Deep structured CNNs have simply more number of layers. Profound CNNs are utilized for classification as well as for object detection. However, this can be extended for face detection and face recognition. CNNs are similar to the conventional neural network. They both have neurons and learnable weights and biases. Every neuron is given an input, performs dot product, and performs nonlinearity. CNNs have multiple layers. Each layer receives a multidimensional array of inputs and produces the multidimensional output, which becomes input to the next layer. Hence, CNN is a sequence of multiple layers. For acquiring the best execution of such a technique, it needs to exceptionally tune the number of nodes, layers, and rates for learning. Some practical applications are included for image classification and face detection.