Semantic Image Classification using Deep Neural Network with Bag of Words layer

The Bag of Visual Words (BoVWs) model is an important concept in computer vision that can be used to classify an image using image features as visual words. It represents image as a collection of unordered image patches. It has three independent steps, feature detection, feature description, and codebook generation and they are hard to be joined. It uses local features, it leads to semantic gap. To address all these issues, in our model Bag of Words layer is used in deep neural network to generate category specific visual words. Deep neural network has convolutional layers that extract more features so it also helps to solve semantic gap issue. By this technique CBIR becomes an easy one. The proposed model will retrieve similar images based on the user query image, by analyzing the features or content of the given image. It also uses the inverted file index strategy for image retrieval process. Fast convergence of training procedure, semantically discriminative ability and sparsity of visual words are added to achieve high performance of classification and retrieval process.