Adaptive classification of artistic images using multi-scale convolutional neural networks.

Area of Science:

• Computer Science
• Artificial Intelligence
• Image Processing

Background:

• Current art image classification methods suffer from low recall and accuracy.
• Effective classification of art images is crucial for digital archives and analysis.

Purpose of the Study:

• To enhance the performance of art image classification.
• To develop a novel adaptive classification method using multi-scale convolutional neural networks (CNNs).

Main Methods:

• Employed multi-scale Retinex algorithm with color recovery for image enhancement.
• Utilized extreme pixel ratio for image quality evaluation and edge detection for feature extraction.
• Constructed a multi-scale CNN with dilated convolutions and employed decision fusion for classification.

Main Results:

• The proposed method significantly improved recall and precision rates for art image classification.
• Achieved reliable and accurate classification results for diverse art images.

Conclusions:

• The adaptive multi-scale CNN method effectively addresses limitations in existing art image classification.
• This approach offers a robust solution for accurate and reliable art image categorization.