Learning Deep Blind Quality Assessment for Cartoon Images.

Area of Science:

• Computer Vision
• Image Processing
• Artificial Intelligence

Background:

• The rapid growth of the cartoon industry highlights a gap in specialized image quality assessment (IQA) methods.
• Existing natural scene IQA algorithms perform poorly on cartoons, failing to align with human perception.
• Manual annotation for training deep learning models is resource-intensive.

Purpose of the Study:

• To develop a robust blind (no-reference) cartoon image quality assessment (IQA) method.
• To overcome the limitations of manual data labeling for training deep learning models.
• To improve the accuracy and reliability of IQA for cartoon images.

Main Methods:

• A novel blind cartoon IQA method leveraging convolutional neural networks (CNNs).
• Development of a full-reference (FR) cartoon IQA metric based on cartoon-texture decomposition.
• Utilizing the FR metric to guide the training of a no-reference IQA network.
• Implementation of a stochastic degradation strategy and a large-scale dataset for enhanced network robustness.

Main Results:

• The proposed method achieves effective and robust performance on both synthetic and real-world cartoon datasets.
• Experimental results demonstrate superior performance compared to existing IQA approaches for cartoons.
• The FR-guided approach significantly improves the accuracy of no-reference IQA.

Conclusions:

• The developed blind cartoon IQA method effectively addresses the limitations of existing algorithms.
• The proposed technique offers a more reliable and perceptually aligned assessment of cartoon image quality.
• This work provides a valuable tool for the cartoon industry to ensure high-quality visual content.