Skin Lesion Segmentation Using Multiple Density Clustering Algorithm MDCUT And Region Growing

• Published in: 2018 IEEE ACIS 17th International Conference on Computer and Information Science (ICIS) pp. 74 - 79
• Main Authors: Louhichi, Soumaya, Gzara, Mariem, Abdallah, Hanene Ben
• Format: Conference Proceeding
• Language: English
• Published: IEEE 01.06.2018
• Subjects: Clustering algorithms; density based clustering; dermoscopic images; Feature extraction; Image color analysis; Image segmentation; Lesions; MDCUT; region growing; segmentation; Shape; Skin; skin lesion
• DOI: 10.1109/ICIS.2018.8466531

Skin lesion segmentation is a key step in a diagnosis system based on dermoscopic images. This paper proposes a method to detect the skin lesion accurately. The images are first cleansed to remove noise. Then, pertinent features are extracted from RGB, HSV and XYZ color spaces. Cluster analysis is used for segmentation. We take advantage of the multiple density clustering algorithm MDCUT [1] to solve the problem of image segmentation using region growing. We demonstrate how MDCUT algorithm is used to automatically determine the needed parameters for region growing image segmentation. Experiments on medical skin lesion image and comparison with the ground truth segmentation results demonstrate the validity of our method.