Machine Learning Based Classification of Multi-Component Shape Descriptors

The concept of Multi-component shapes is motivated by researchers in image classification and computer vision applications. In this paper, we have introduced several descriptors that are modified versions of the Hu moment invariants and suitable for measuring certain properties of multi-component shapes. These descriptors are invariant in translation, rotation, and scaling-based transformation, and this is important because they can significantly develop systems for object identification, classification, matching, and analysis. The classification of multiple objects from the MPEG-7 dataset is applied using the descriptors introduced in this research. Multiple machine learning classifiers are used to test the new descriptors’ performance, where SVM has shown notable results i.e., 87.5%, 88.96, 87.5, and 88.23 for Accuracy, Precision, Recall, and F1-score, respectively, as proof of the worthiness of new descriptors.