Improving shape classification accuracy with Binary Directional Shape descriptors in image processing

In this paper, we introduce a novel Binary Directional Shape (BDS) descriptor for efficient object shape classification. The BDS descriptor is designed to operate in real-time environments, balancing computational efficiency with high classification accuracy. The descriptor works by resizing input images, thresholding them into binary representations, and analyzing them along multiple predefined directions. Configurable parameters such as image size, number of directions, and step size allow for customization based on application needs. Through extensive experiments, we explore the effects of image resolution on shape classification, demonstrating how aliasing and resolution loss can impact performance. Our results show that while higher resolutions improve accuracy, challenges such as rotation variance and aliasing persist. The BDS descriptor outperforms traditional feature-based methods, providing superior classification results even in the presence of these challenges. This work demonstrates the value of directional information in shape recognition and sets the stage for future optimizations and applications.