Initialization of Active Contour for Dermoscopic Image Segmentation: A Comparative Study

摘要

Segmentation is an important step in medical image processing and analysis. Active contour (AC) is the one of the most effective methods for image segmentation. In this paper, the model of active contour without edge is used to segment the lesions in dermoscopic images, where it is more effective than the traditional snake. However, its performance depends on the initial curve. This motivates the current work to apply different shape of mask as an initial contour of Chan-Vese segmentation method to study the segmentation performance on the dermoscopic image. Evaluation metrics are calculated to measure the segmentation performance. The experimental results establish that the circular initial mask provids the best performance, where it is approximately has the same shape of the skin lesion regions. In addition, this initial circular mask achieves the best segmentation performance if it has 49% size from the whole dermoscopic image size.