Reproduce “Integrated Neural Network and Machine Vision Approach For Leather Defect Classification” on OpenResearch

About this paper

Sze-Teng Liong, Y.S. Gan, Yen-Chang Huang, Kun-Hong Liu, Wei-Chuen Yau

Liong et al. developed an integrated approach for automatic classification of tick-bite defects on calf leather, combining handcrafted feature extraction with Artificial Neural Networks. The ApproxCanny edge detector achieved 83.50% accuracy, while the ANN demonstrated competitive performance at 82.49%.

Key insights

• Well-designed handcrafted features, specifically the ApproxCanny edge detector, can be competitive with Artificial Neural Network (ANN) approaches for certain image classification tasks, slightly outperforming the best ANN configuration in this study.
• Different feature extraction methods, such as edge detection and statistical approaches, capture complementary aspects of leather texture and defects.
• Dataset imbalance significantly impacts classification performance, especially for the minority class, highlighting the need for strategies to address skewed data distributions in real-world applications.

Summary by alphaXiv.