ICIP 2023 CHALLENGE: FULL-REFERENCE AND NON-REFERENCE POINT CLOUD QUALITY ASSESSMENT METHODS WITH SUPPORT VECTOR REGRESSION

In recent years, point clouds have gained much interest in a variety of 3-dimensional (3-D) applications, for instance, 3-D immersive telepresence. In this application, point clouds must be compressed for transmission which introduces several types of distortion. Thus, point cloud quality assessment is an essential component needed to evaluate the point cloud quality for a given application. Based on this background, ICIP 2023 Point Cloud Visual Quality Assessment Grand Challenge (ICIP 2023 PCVQA) is organized to establish better point cloud quality assessment methods. In this challenge, an assessment scheme is categorized into full-reference (FR) and non-reference (NR). In this paper, we describe our proposed metrics for both the FR and NR categories. Although both methods are based on support vector regression (SVR) to obtain the assessment score, the features utilized to train the SVR model are different. For the FR metric, 5 types of FR scores including geometric and color distortion are utilized to train an SVR model. As for the proposed NR metric, the features are acquired by a graph constructed by a point cloud. Our experimental results show that the proposed metrics outperform recent state-of-the-art point cloud quality assessment methods.