Digital Microscopic Image Sensing and Processing for Leather Species Identification

• Published in: IEEE sensors journal Vol. 20; no. 17; pp. 10045 - 10056
• Main Authors: Varghese, Anjli, Jain, Sahil, Prince, A. Amalin, Jawahar, Malathy
• Format: Journal Article
• Language: English
• Published: New York IEEE 01.09.2020; The Institute of Electrical and Electronics Engineers, Inc. (IEEE)
• Subjects: Algorithms; Automatic species identification; Automation; circular Hough transform (CHT); Cost analysis; Cows; Data processing; Digital imaging; digital microscopic image sensing; Digital signatures; Fraud; Hair; hair-pore segmentation; Hough transformation; Identification; Image acquisition; Image analysis; Image segmentation; k-nearest neighbor (KNN); Leather; leather image data; Machine learning; Microscopy; morphological features; Morphology; Multilayers; Pattern recognition; Sensors; Skin; Surface morphology
• ISSN: 1530-437X; 1558-1748
• DOI: 10.1109/JSEN.2020.2991881

Leather is a durable material well-known for its fashion, style, and versatility. Identifying the animal species from which leather originated is necessary in leather quality-check, fraud detection, exotic animal protection, etc. The species identification techniques currently in practice involve subjective and supervised analysis with laboratory-specific devices. This paper discusses optimized and automated species identification by employing a portable and cost-effective (economically efficient) digital microscope. The goal is to acquire the leather images of the four most predominantly used permissible species, with the definite hair-pore regions. Preliminary experiments investigate the adequate image sensing parameters for efficient sensor data processing. Otsu's thresholding followed by circular Hough transform (CHT) segments and estimates the morphological features of the informative hair-pore regions. The k-nearest neighbor (KNN) based machine learning algorithm models a pattern recognition technique for automated species prediction. Evaluation measures objectively validate the performance of the proposed pre-processing and hair-pore segmentation. The experimental analysis presents the uniqueness and significance of estimated morphological features. The study also compares KNN and Multi-Layer Perceptron (MLP) based species prediction. The comparative analysis ascertains the significance of KNN-based leather species identification with 92.5% accuracy. Thus, the present research assists in building the digital signatures of permissible leather species. It also contributes to design a cost-effective and automated leather species prediction technique with objective analysis.