This paper presents a deep neural network-based approach for full-reference (FR) and no-reference (NR) image quality assessment (IQA). The network is trained end-to-end with 10 convolutional layers, 5 pooling layers for feature extraction, and 2 fully connected layers for regression, making it significantly deeper than existing IQA models. The proposed architecture can be adapted for both FR and NR IQA settings and allows for joint learning of local quality and local weights, i.e., the relative importance of local quality to the global quality estimate. The approach is purely data-driven and does not rely on hand-crafted features or prior domain knowledge about the human visual system or image statistics. The model is evaluated on the LIVE, CISQ, and TID2013 databases, as well as the LIVE In the Wild Image Quality Challenge Database, showing superior performance to state-of-the-art NR and FR IQA methods. Cross-database evaluation demonstrates the model's ability to generalize between different databases, indicating high robustness of the learned features. The paper discusses the use of deep convolutional neural networks (CNNs) in IQA, highlighting their ability to outperform traditional approaches by allowing end-to-end learning of features and regression. The proposed method uses a CNN architecture inspired by the organization of the primate visual cortex, comprising 10 convolutional layers and 5 pooling layers for feature extraction, and 2 fully connected layers for regression. The model is adapted for both FR and NR IQA by modifying the feature extraction paths and introducing a feature fusion step to allow for joint regression of features extracted from the reference and distorted image. The model is evaluated on various IQA databases, showing superior performance compared to existing methods. The paper also discusses the importance of spatial pooling in IQA, with weighted average patch aggregation showing improved performance in some cases. The proposed method is able to adapt to different distortion types and is shown to perform well on challenging databases such as CLIVE. The results indicate that the proposed method is effective for both FR and NR IQA and has the potential to be used in a wide range of applications.This paper presents a deep neural network-based approach for full-reference (FR) and no-reference (NR) image quality assessment (IQA). The network is trained end-to-end with 10 convolutional layers, 5 pooling layers for feature extraction, and 2 fully connected layers for regression, making it significantly deeper than existing IQA models. The proposed architecture can be adapted for both FR and NR IQA settings and allows for joint learning of local quality and local weights, i.e., the relative importance of local quality to the global quality estimate. The approach is purely data-driven and does not rely on hand-crafted features or prior domain knowledge about the human visual system or image statistics. The model is evaluated on the LIVE, CISQ, and TID2013 databases, as well as the LIVE In the Wild Image Quality Challenge Database, showing superior performance to state-of-the-art NR and FR IQA methods. Cross-database evaluation demonstrates the model's ability to generalize between different databases, indicating high robustness of the learned features. The paper discusses the use of deep convolutional neural networks (CNNs) in IQA, highlighting their ability to outperform traditional approaches by allowing end-to-end learning of features and regression. The proposed method uses a CNN architecture inspired by the organization of the primate visual cortex, comprising 10 convolutional layers and 5 pooling layers for feature extraction, and 2 fully connected layers for regression. The model is adapted for both FR and NR IQA by modifying the feature extraction paths and introducing a feature fusion step to allow for joint regression of features extracted from the reference and distorted image. The model is evaluated on various IQA databases, showing superior performance compared to existing methods. The paper also discusses the importance of spatial pooling in IQA, with weighted average patch aggregation showing improved performance in some cases. The proposed method is able to adapt to different distortion types and is shown to perform well on challenging databases such as CLIVE. The results indicate that the proposed method is effective for both FR and NR IQA and has the potential to be used in a wide range of applications.