研究目的
To establish an effective method to measure the quality of stereoscopic images automatically using a blind image quality assessment model that combines classification and prediction stages.
研究成果
The proposed CAP-3DIQA model significantly improves the performance of stereoscopic image quality assessment, particularly for asymmetrically distorted images, and shows better generalization in cross-database evaluations compared to existing methods. It combines hierarchical classification with multi-channel prediction to mimic human visual perception effectively.
研究不足
The classification accuracy is not perfect, especially for certain distortion types, and the model may not perform as well on symmetrically distorted images compared to asymmetrically distorted ones. The method relies on specific databases and may not generalize to all types of distortions or images.
1：Experimental Design and Method Selection:
The proposed model, CAP-3DIQA, uses a hierarchical learning approach with classification and prediction stages. It employs support vector classification (SVC) for distortion type classification and support vector regression (SVR) and random forest (RF) for quality score prediction.
2：Sample Selection and Data Sources:
Three public databases are used: LIVE 3D Image Quality Database Phase I, LIVE 3D Image Quality Database Phase II, and MCL 3D Image Quality Database, containing stereoscopic image pairs with various distortions.
3：List of Experimental Equipment and Materials:
No specific equipment is mentioned; the method is computational and uses software tools like LIB-SVM.
4：Experimental Procedures and Operational Workflow:
Distorted images are classified into subsets based on distortion types using hierarchical binary classifiers. Then, five prediction channels (left-view, right-view, and three binocular visual combination images) extract features (e.g., LBP, SVD) and predict quality scores, which are fused using SVR.
5：Data Analysis Methods:
Performance is evaluated using Pearson linear correlation coefficient (PLCC), Spearman rank ordered correlation coefficient (SROCC), and root mean squared error (RMSE), with 1000 iterations of 80-20 train-test splits.
独家科研数据包，助您复现前沿成果，加速创新突破
获取完整内容
