Deep neural network as deep feature learner

Features play an important role in image processing. But as not all features are comparable, relative features emerged. From the beginning, low-level features, extracted by experts, have been employed to create difficult models for learning the problem of relative attribute. Knowing these models are limited in generality of their applicability, deep learning models can be employed instead of them. A deep artificial neural network framework has been suggested for the task of relative attribute prediction in this article. The paper suggests to use a convolutional artificial neural network for learning the mentioned attributes through a peripheral auxiliary layer, called also a ranking layer, which is able to learn how to rank the images. A suitable ranking cost function is used to train the whole network in an end-to-end manner. The suggested method through this paper is experimentally superior to the state of the art methods on some well-known benchmarks. The experimental results indicate that the proposed method is capable of learning the problem of relative attribute.