Maximum Entropy Based Color Image Segmentation with Probabilistic Rand Index

Image segmentation is an important part of image processing, and it refers to the process of partitioning a digital image into multiple meaningful segments. In this project thesis, we use the Berkeley Segmentation Database to test our segmentation scheme and evaluate it against the multiple ground truth segmentations provided by different human subjects. This work proposes a color image segmentation approach and evaluates segmentation quality using the Probabilistic Rand Index (PRI). The Probabilistic Rand Index (PRI) is used to measure the similarity between our segmented outputs and the corresponding ground truth segmentations. All segmentation experiments and evaluations were implemented in MATLAB. To understand the effectiveness of the proposed method, we also compare our PRI results with those reported in a previous article that presented a multi-objective segmentation strategy for the same color image indices. Based on this comparison, our approach achieves the best PRI for the maximum number of image indices in the test set. Due to time constraints, the same procedure could not be extended to medical images within the scope of this thesis.

• Research Project