This paper presents a deep neural network-based approach for image quality assessment (IQA), which can be used in both no-reference (NR) and full-reference (FR) settings. The network, consisting of 10 convolutional layers and 5 pooling layers for feature extraction, and 2 fully connected layers for regression, is significantly deeper than most related IQA models. Key features of the proposed architecture include its ability to handle both NR and FR IQA settings and its capability to jointly learn local quality and local weights in a unified framework. The approach is purely data-driven, relying on raw input data without hand-crafted features or domain knowledge about the human visual system. The performance of the proposed method is evaluated on several databases, including LIVE, CISQ, TID2013, and the LIVE In the Wild Image Quality Challenge Database, showing superior performance compared to state-of-the-art NR and FR IQA methods. Cross-database evaluation demonstrates the method's ability to generalize across different datasets, indicating robustness in learned features. The paper also discusses the impact of network depth, feature fusion strategies, and the number of patches used on the performance of the proposed method.This paper presents a deep neural network-based approach for image quality assessment (IQA), which can be used in both no-reference (NR) and full-reference (FR) settings. The network, consisting of 10 convolutional layers and 5 pooling layers for feature extraction, and 2 fully connected layers for regression, is significantly deeper than most related IQA models. Key features of the proposed architecture include its ability to handle both NR and FR IQA settings and its capability to jointly learn local quality and local weights in a unified framework. The approach is purely data-driven, relying on raw input data without hand-crafted features or domain knowledge about the human visual system. The performance of the proposed method is evaluated on several databases, including LIVE, CISQ, TID2013, and the LIVE In the Wild Image Quality Challenge Database, showing superior performance compared to state-of-the-art NR and FR IQA methods. Cross-database evaluation demonstrates the method's ability to generalize across different datasets, indicating robustness in learned features. The paper also discusses the impact of network depth, feature fusion strategies, and the number of patches used on the performance of the proposed method.