Point cloud quality assessment (PCQA) is a critical research area focused on evaluating the perceptual Quality of Experience (QoE) of point clouds to enhance visual experiences of immersive multimedia applications for end users. To prevent the complex computations on 3D data applied by model-based methods, projection-based models have been developed to estimate the QoE by analysing 2D projection views of the point cloud. In this paper, we propose a novel projection-based No-Reference (NR) PCQA method, called Multi-View Adaptive Weighting Point Cloud Quality Assessment (MVAW-PCQA), to predict the QoE of distorted point clouds using six 2D projection views as the input of a convolutional neural network (CNN) architecture. First, multi-view involves independently extracting features from multiple projection views of a point cloud, guaranteeing view-specific features are learned without prematurely mixing spatial information, and preserving the unique contributions of each projection view to the final quality prediction. Then, an adaptive weighting fusion mechanism combines the features extracted from the different projection views by learning their relative importance. This design enables the model to focus on the most informative projections for predicting the point cloud quality. The experimental results demonstrate that our method outperforms state-of-the-art NR-PCQA methods on the SJTU-PCQA dataset in terms of root mean square error (RMSE) and correlation coefficients (Pearson, Spearman, and Kendall), while adopting a lightweight design with a reasonable number of parameters for the trained neural network.

This paper reviews the VCIP 2025 Grand Challenge on Live Broadcasting Video Quality Assessment. The competition aims to foster innovation in both subjective and objective VQA techniques tailored to live broadcasting videos, addressing the unique challenges posed by live streaming impairments while emphasizing the evaluation of QoE. The grand challenge used live broadcasting database LBVD which consists of 1013 videos focusing on distortion in live broadcasting videos. The competition had 14 participants and 5 teams submitted valid solutions for the final testing phase. The proposed solutions have shown significant progress in areas such as combining traditional feature engineering with deep learning models, achieved state-of-the-art performances for LBVD. Team ATHENA-Live-QoE and Team HZX Force tied for the first position. The dataset can be found at https://github.com/cpf0079/LBVD.

Ensuring high-quality multimedia experiences at scale requires objective models of Quality of Experience (QoE) that approximate human perception reliably across diverse contents, devices, and networks. This doctoral thesis addresses the problem of defining and deploying such models by proposing a unified methodology structured around five principles: careful signal design, temporal modeling of event sequences, multi-view deep learning fusion under data constraints, perceptual alignment through thresholds such as JND/SUR, and efficiency for deployment in resource-limited settings. The methodology has been instantiated across multiple studies, ranging from real-time communication and adaptive streaming to point cloud and face image quality assessment. Across these use cases, the models show strong agreement with subjective judgments, robustness to content and device shifts, and efficiency compatible with online deployment. Collectively, these contributions establish a general recipe for perception-aware QoE modeling, bridging idea and operational practice.

Just Noticeable Difference (JND) thresholds, particularly for quality metrics such as Video Multimethod Assessment Fusion (VMAF), are critical in streaming, helping identify when quality changes become perceptible and reducing redundant bitrate representations. The Satisfied User Ratio (SUR) complements JND by quantifying the percentage of users who do not perceive a difference, offering practical guidance for perceptually optimized streaming. This paper proposes a novel two-branch deep neural network (DNN) for predicting the 75% SUR for VMAF, the encoding level where 75% of viewers cannot perceive degradation. The framework combines handcrafted features (e.g., spatial and temporal indicators such as SI, TI, etc.) and deep learning-based (DL-based) representations extracted via a convolutional neural network (CNN) backbone. The DL-based branch employs a spatio-temporal attention mechanism and a Long Short-Term Memory (LSTM) to capture temporal dynamics, while the handcrafted branch encodes interpretable indicators through a fully connected layer. Both outputs are fused and passed through a lightweight Multilayer Perceptron (MLP) to predict 75% SUR. To improve robustness to noise and label uncertainty, the model is trained using the Smooth-L1 loss. Experiments on the VideoSet dataset show our method outperforms SOTA across all metrics, achieving a notably higher R2score (0.46 vs. 0.36), indicating improved prediction reliability and low computational complexity, making it suitable for real-time video streaming.

Multi-view (MV) learning is a machine learning technique for improving generalization efficiency by learning from different feature subsets derived from multiple sources. We believe this approach can help in Quality of Experience (QoE) modelling by integrating knowledge from different datasets generated by subjective tests conducted for the same or similar applications considering different QoE Influence Factors (IFs). To investigate this subject, in this paper, we present the experiments conducted starting from a complete dataset related to Web browsing sessions that has been artificially divided into two distinct subsets (views). The proposed MV learning approach implements a data fusion technique to integrate extracted features from different views into a unified feature space. To achieve a complete experiment on the entire problem space, all possible combinations of IFs (features) in two distinct partial views (PVs) are considered and trained in the MV approach; the full view (FV) approach, which utilizes the complete dataset, is also considered for performance comparison. Experimental results show the QoE estimation performance achieved by the MV (0.69) is comparable with that of the FV (0.72), although the 2 single views were used for training in the MV case. Moreover, the performance enhancement achieved by the MV compared with the PV is most noticeable when a lower number of features is used to train the models.