Texture and Shape-Based Segmentation of Coffee Beans Images

This study introduces an innovative approach to the challenge of segmenting coffee bean images, employing a Bayesian classification method that leverages both texture and shape attributes. The core of our system is its novel exploration of texture characteristics across different ripeness stages of coffee beans, enriched by shape descriptors to address and rectify the issues of sub-segmentation. These challenges are particularly pronounced due to the complex nature of real-world images, characterized by overlapping, merging, or occlusion among beans. Additionally, this paper outlines a vision for implementing the proposed methodology within an efficient hardware architecture, aiming for practical application in the near future. We present segmentation outcomes using real-world images, conducting thorough comparisons with other segmentation techniques. This includes a comprehensive evaluation of the system’s computational complexity and memory requirements, showcasing the effectiveness and efficiency of our proposed solution.