The control and monitoring of the distribution, morphology, and interactions of granular objects with regular and irregular shapes remain a major challenge in many scientific and industrial fields. Automatic analysis of such objects from 2D images is particularly difficult in dense scenes characterized by small particles, object contacts, and varying illumination conditions. The objective of this study is to develop and evaluate different image processing and machine learning approaches for reliable detection and segmentation of granular objects placed on a planar surface. The analysis focuses on the extraction of quantitative indicators such as surface coverage rate, object count, and the proportion of objects in contact. In a first stage, a classical segmentation method based on HSV color-space thresholding, as well as a segmentation model applied without task-specific training, are implemented and compared in order to analyze their performance and limitations, revealing difficulties in separating adjacent objects and handling illumination variations. In a second stage, these methods are replaced by task-specific deep learning–based segmentation models, namely YOLOv8-Seg and a fine-tuned version of the Segment Anything Model (SAM), trained using an annotated dataset. Performance is evaluated using standard computer vision metrics, including Intersection over Union (IoU), Dice coefficient, and detection precision. The results demonstrate that the trained models, especially the fine-tuned SAM, yield enhanced segmentation and superior separation of adjacent objects, facilitating more dependable quantitative analysis of intricate granular environments.