A Novel Model for Visual Content Based Image Retrieval using Transfer Learning

Abstract

At present, the revolution brought by deep learning based technologies in the field of computer vision gaining momentum in the world of artificial intelligence. In particular, the best models for retrieving common images today are based on features generated by deep convolutional neural networks (DCNNs). However, this great success was expensive. A comprehensive amount of tagged data had to be collected, followed by model design and training. Meanwhile, a transfer-of learning approach has been developed that avoids this costly step by applying a sophisticated, pre-trained generic DCNN model to completely different data domains. With the use of transfer learning, it becomes possible to use deep CNN models for small datasets with better retrieval performance with respect to handcrafted feature based retrieval methods. In this paper a deep CNN based model has been proposed which uses concept of transfer learning and achieves good classification accuracy.