Image aesthetic quality assessment: A method based on deep convolutional capsule network.

Area of Science:

• Computer Vision
• Artificial Intelligence
• Machine Learning

Background:

• Image aesthetics assessment (IAA) is a growing field with significant application potential.
• Current IAA methods often neglect crucial spatial information, limiting aesthetic evaluation accuracy.
• There is a need for advanced models that can effectively capture spatial relationships for better image quality assessment.

Purpose of the Study:

• To propose a novel method, the Deep Convolutional Capsule Network (DCCN), for image aesthetics assessment.
• To enhance the representation of spatial features in IAA by integrating capsule networks.
• To improve the accuracy and robustness of automated image aesthetic evaluation.

Main Methods:

• Developed a Deep Convolutional Capsule Network (DCCN) integrating an improved Inception module with capsule routing.
• The DCCN is designed to extract both global and local aesthetic features while preserving spatial relationships.
• The proposed method was evaluated on the CUHK-PQ and AVA benchmark datasets.

Main Results:

• The DCCN achieved 94.79% classification accuracy on the CUHK-PQ dataset.
• On the AVA dataset, the DCCN obtained a Pearson Linear Correlation Coefficient (PLCC) of 0.8408 and a Spearman Rank-Ordered Correlation Coefficient (SROCC) of 0.7394.
• The results demonstrate the effectiveness of incorporating capsule networks for spatial feature representation in IAA.

Conclusions:

• The Deep Convolutional Capsule Network (DCCN) represents a novel and effective approach to image aesthetics assessment.
• The method successfully enhances spatial feature extraction, leading to improved performance on benchmark datasets.
• Future work should address the DCCN's sensitivity to style variations, resolution changes, and inference complexity for real-time applications.