PROJECT TITLE :

End-to-End Blind Image Quality Assessment Using Deep Neural Networks - 2018

ABSTRACT:

We have a tendency to propose a multi-task finish-to-end optimized deep neural network (MEON) for blind image quality assessment (BIQA). MEON consists of 2 sub-networks-a distortion identification network and a high quality prediction network-sharing the early layers. In contrast to traditional methods used for coaching multi-task networks, our training method is performed in two steps. In the first step, we have a tendency to train a distortion kind identification sub-network, for that giant-scale coaching samples are readily offered. Within the second step, starting from the pre-trained early layers and the outputs of the first sub-network, we train a quality prediction sub-network employing a variant of the stochastic gradient descent method. Completely different from most deep neural networks, we have a tendency to choose biologically galvanized generalized divisive normalization (GDN) instead of rectified linear unit because the activation function. We tend to empirically demonstrate that GDN is effective at reducing model parameters/layers while achieving similar quality prediction performance. With modest model complexity, the proposed MEON index achieves state-of-the-art performance on four publicly on the market benchmarks. Moreover, we have a tendency to demonstrate the sturdy competitiveness of MEON against state-of-the-art BIQA models using the cluster most differentiation competition methodology.