Avaliação Objetiva de Qualidade de Imagem e Vídeo 3D Utilizando Técnicas de Aprendizado de Máquina

Abstract

Decades of research in Image and Video Quality Assessment have led to the creation of a variety of objective quality metrics that strongly correlate with subjective image and video quality. However, challenges remain when considering quality assessment of 3D/stereo images and videos. Multiple objective quality metrics for 3D images and videos were designed by extending the 2D metrics. As a result, these solutions tend to present limitations associated with 3D-specific artifacts. Recent work has demonstrated the effectiveness of machine learning techniques in developing 3D quality metrics. While effective, some machine learning solutions demand high computational effort restricting their adoption for low latency applications and/or embedded systems. In this context, a set of important research questions arise justifying this Thesis. We present a study on the evaluation of full-reference stereo scopic objective quality considering lightweight machine learning. In this study we discretized the quality score in order to treat the quality assessment as a classification problem. We evaluated four different algorithms based on decision trees considering eight different sets of image/video characteristics. In addition, we performed tests considering different numbers of classes for both image and video quality evaluation. The classifiers were trained using data from the Waterloo IVC 3D Image Quality Database and Waterloo IVC 3D Video Quality Database to determine the subjective quality score measured using the Mean Opinion Score (MOS). The results show that, in general, RandomForest presents the best accuracy. Our study demonstrates the feasibility of decision tree solutions as an effective and lightweight approach for assessing 3D image and video quality.