Directional local region-scalable active contour using expandable kernel
Faisal Amir, Charnchai Pluempitiwiriyawej · 2012
This paper extracts another benefit of local region-scalable active contour using expandable kernel (LREK) [1]. We modify it to posses a directional property so that it is capable in detecting objects of desirable edge's type. Local region-scalable force is determined by pixel's intensity profile within expandable kernels centered on the contour. While magnitude of intensity difference is used in adaptive local statistics of expandable kernel, sign of the difference is a directional information in controlling the forces to attract the contour only into a particular edge's type object. By setting a directional parameter, an initial contour placement results in two different segmented outcomes. Formulated via level set, our model is topologically flexible. Local region-scalable information enables our model to converge quickly into desired objects with noises, intensity inhomogeneity, and heterogeneous textures. Moreover, the adaptive local statistics allows in tracing concave boundary with a large capture range.