Image Segmentation by Hybrid Active Contour Model Using Harmonic Mean
Irfana Bibi,Liu Fang,Amir Razi,Yuanhao Cui
DOI: https://doi.org/10.1109/icisce.2017.83
2017-01-01
Abstract:Edge-based models can identify the objects having solid borderline effectively, however, these models cannot appropriately segment the images with weak object borderline. On the other hand, region-based models perform better for images with unclear object borderline, however, these models cannot accurately segment images with intensity inhomogeneity. In this study, we propose a hybrid Active Contour Model (ACM) that can overcome these limitations. It is formed by substituting the edge information into region-based level set with a harmonic mean based signed pressure force function HSPF. Due to the utilization of harmonic mean intensities of the image in HSPF, the new data fitting energy can better uphold the subtle difference among the desired object intensities and that of its background that results in the effective segmentation of low contrast images. Since our proposed model has both, the region and edge information, so in comparison to traditional models, it performs better in the segmentation of images with regions of interest (ROI) having noise, indistinct borders, and intensity inhomogeneity. The proposed model is tested on noisy, real and medical images to exhibit its correctness, efficiency, and robustness. The experimental results validate that our proposed model is more efficient, rapid and correct as compared to the other related models.