Full Reference Mesh Visual Quality Assessment Using Pre-Trained Deep Network and Quality Indices

In this paper, we propose an objective quality metric to evaluate the perceived visual quality of 3D meshes. Our method relies on pre-trained convolutional neural network i.e VGG to extract features from the distorted mesh and its reference. Quality indices from well-known mesh visual quality metrics are concatenated with the extracted features resulting a global feature vector. this latter is used to learn the support vector regression (SVR) to predict the final quality score. Experimental results from two subjective databases (LIRIS masking database and LIRIS/EPFL general-purpose database) and comparisons with seven objective metrics cited in the state-of-the-art demonstrate the effectiveness of the proposed metric in terms of the correlation to the mean opinion scores across these databases.