Automatic Gemstone Classification Using Computer Vision

This paper presents a computer-vision-based methodology for automatic image-based classification of 2042 training images and 284 unseen (test) divided into 68 categories gemstones. A series feature extraction techniques (33 including colour histograms in the RGB, HSV CIELAB space, local binary pattern, Haralick texture grey-level co-occurrence matrix properties) were used combination with different machine-learning algorithms (Logistic Regression, Linear Discriminant Analysis, K-Nearest Neighbour, Decision Tree, Random Forest, Naive Bayes Support Vector Machine). Deep-learning ResNet-18 ResNet-50 was also investigated. The optimal provided by Forest algorithm RGB eight-bin histogram pattern features, an accuracy 69.4% on images; required 0.0165 s to process test images. These results compared against three expert gemmologists at least 5 years experience gemstone identification, who obtained accuracies between 42.6% 66.9% took 42–175 min classify As expected, human experts much longer than computer vision algorithms, which addition provided, albeit marginal, higher accuracy. Although these experiments included relatively low number images, superiority over humans is line what has been reported other areas study, it encouraging further explore application gemmology related areas.