Abstract:
Stereoscopic imaging is becoming increasingly popular, and its use in photography, television, and films is rapidly expanding. Obviously, access to this type of images often includes necessary treatments (acquisition, processing, compression, transmission, etc.), which may result in a variety of artifacts (blocking, blur, ringing, etc.). As a result, it is critical to have adequate tools for measuring the quality of stereoscopic contents. It is thus essential to establish efficient metrics that assess the impact of these treatments on the perceived quality. To meet this critical need, significant efforts have been made to study and evaluate the quality of stereoscopic images. In this thesis, we present several contributions for quality assessment of stereoscopic contents. Five methods have been proposed in total, with all of them are no-reference based metric. These metrics were developed with Human Visual System (HVS) modeling and human visual attention (saliency information) in mind. In addition, various advanced techniques, such as deep learning, have been incorporated into our workflow designs.
