Blind Stereoscopic Image Quality Assessment Using Convolutional Neural Networks and Support Vector Regression

Abstract

Image quality presents an important aspect of several applications such as biometrics, tracking, object detection and so on. To estimate automatically the quality, several measures have been proposed in the literature. These measures aim to predict the subjective judgment of a given image according to different characteristics. The overarching of this paper is to present a framework for blind stereoscopic image quality metric based on Convolutional Neural Network (CNN) and Support Vector Regression (SVR). The proposed CNN model is composed of 3 convolutional layers and two Fully Connected (FC) layers and, aims to identify the degradation type in the image. The quality is then estimated using an SVR model whose inputs are some computed features, selected according to the identified degradation type. The obtained results through two common datasets show the relevance of the proposed approach.