Image Embedding and Classification using Pre-Trained Deep Learning Architectures

Abstract

The use of deep learning in image analysis has been a game-changer. Common methods have relied on problem-specific algorithms to characterize pictures using properties like cell shape, count, intensity, and texture that were painstakingly constructed by human experts. With deep convolutional neural networks, feature learning occurs implicitly, and training the network often zeroes down on specific tasks. To get a vector representation of a picture by embedding, the image is fed into a preexisting deep neural network. An image embedding is a representation of the picture in fewer dimensions. It is a dense vector representation of the picture that may be used for things like categorization and plenty of other things. In this research, we profile pictures by using vectors of features and pre-trained deep learning architectures (Inception V3, VGG16, VGG19, Painter, SqueezeNet, and DeepLoc). The process of picture categorization makes use of these vectors. For experimentation purposes, we used the dataset containing images of yoga positions and classified them using Multilayer Perceptron (MLP) Classifier.