Design of Chinese painting style classification model based on multi-layer aggregation CNN

Area of Science:

• Computer Vision
• Artificial Intelligence
• Art History

Background:

• Traditional Chinese Painting (TCP) possesses significant artistic value, yet its emotional dimensions are complex to quantify.
• Bridging the gap between TCP's characteristics and human emotional interpretation requires advanced analytical tools.

Purpose of the Study:

• To develop a novel convolutional neural network (CNN)-based classification model for recognizing emotions in traditional Chinese paintings (TCP).
• To establish a comprehensive framework for analyzing and mapping emotional features within TCP.
• To enhance the understanding and computational analysis of emotional expression in cultural art forms.

Main Methods:

• A novel CNN model was designed, integrating a multi-layer aggregation recalibrated emotional feature module.
• Emotional feature regions were extracted using image saliency.
• Multi-category weighted activation localization was employed for classification within the CMYK color space.

Main Results:

• The proposed CNN model achieved a high accuracy of 92.36% for TCP emotion recognition.
• The algorithm demonstrated superior performance compared to established models like VGG-19, GoogLeNet, ResNet-50, and WSCNet.
• The largest error rate observed was 0.191, indicating robust predictive capability.

Conclusions:

• The developed CNN model effectively captures and interprets the emotional nuances present in traditional Chinese paintings.
• This research advances the integration of multimedia technology with cultural education by providing a quantitative approach to art analysis.
• The findings pave the way for deeper computational understanding of art and emotion.