Benchmarking a New Dataset of Traditional Balinese Carving Ornaments for Image Classification Task

In the framework of the development of an automatic Balinese carving ornament recognition application, a valid image dataset is needed. This paper describes the improved new dataset of traditional Balinese carving ornaments and presents the benchmarking results in an image classification task. The improvement of the new dataset involves the increased number of image samples and involves the validation and addition of the number of ornament classes. Some frequently used feature extraction methods, for example, Gabor Filter, Zoning, Histogram of Gradient, Neighborhood Pixels Weights, and Kirsch edge, were tested to benchmark the image classification task for this new dataset. The benchmark results showed that this new dataset has a fairly high technical challenge for feature extraction methods in the pattern recognition field. The new proposed dataset will support further research steps in building a classification and recognition system for Balinese carving ornaments.