A SIMPLE HYBRID IMAGE SEGMENTATION APPROACH COMBINING CLASSICAL AND DEEP LEARNING TECHNIQUES FOR MEDICAL AND GENERAL IMAGE ANALYSIS

Abstract

Image segmentation is a pivotal step in image processing, enabling precise detection and recognition of objects within  a scene, especially in medical imaging like MRI brain scans. While classical algorithms such as GrabCut offer efficient foreground extraction and graph cut optimization, state-of-the-art deep learning models like Mask R-CNN, U-Net,  FCN, and DeepLab v3 significantly enhance segmentation accuracy and robustness. This paper proposes a simple yet effective hybrid segmentation approach that integrates the strengths of these classical and deep learning methods. Starting with GrabCut for initial foreground extraction, followed by deep learning-based refinement, the framework efficiently handles complex scenes and medical images with limited annotated data. Experimental results on benchmark datasets demonstrate notable improvements in accuracy, recall, F1 score, and boundary precision compared to single- method usage. The proposed technique is easy to implement, adaptable to multiple domains, and offers reliable segmentation performance for both general and medical imaging tasks.