Active contour driven by scalable local regional information on expandable kernel

Amir Faisal (Department of Biomedical Engineering, Institut Teknologi Sumatera, Lampung Selatan, Indonesia, 35365)
Charnchai Pluempitiwiriyawej (Department of Electrical Engineering, Chulalongkorn University, Bangkok, Thailand, 10330)

Journal of Science and Applicative Technology
Vol 4 No 1 (2020): Journal of Science and Applicative Technology June Chapter

An active contour that uses the pixel’s intensity on a set of expandable kernels along the propagating contour for image segmentation is presented in this paper. The objective is this study is to employ the scalable kernels to attract the contour to meet the desired boundary. The key characteristics of this scheme is that the kernels gradually expand to find an object’s boundary. So this scheme could penetrate to the concave boundary more effective and efficient than some other schemes. If a Gaussian kernel is applied, it could trace the object with a blurred or smooth boundary. Moreover, the directional selectivity feature enables in capturing two edge’s types with just one initial position. Its performance showed more desirable segmentation outcomes compared to the other existing active contours using regional information when segmenting the noisy image and the non-uniform (or heterogeneous) textures. Meanwhile, the level set implementation enables topological flexibility to our active contour scheme.