A Comparative Study of Clustering Validation Indices and Maximum Entropy for Sintonization of Automatic Segmentation Techniques

摘要: Automatic image segmentation is a fundamental task in many applications such as video surveillance, retrieval, medical analysis, recognition, tracking and objects classification. This not easy due to the complexity of features well number within images which unknown most time. However, correct tuning parameters associated with automatic algorithms can improve levels precision object images. paper proposes use clustering validation indices maximum entropy cost functions quantify quality order find optimal techniques. The impact using was evaluated algorithms, K-Means, Watershed Statistical Region Merging (SRM), four databases containing different numbers sizes illumination. results obtained reveal that for better segmentations segmented close what really exists images, while these are competitive works reported literature.