A Parallel Realization of the Active Contour Model on Boundary Extraction

Active contour is an effective technique for object boundary extraction in biomedical images such as blood vessel tracing or optic disc extraction in retinal images. The traditional serial realization of contour finding needs considerable processing time. This paper presents a parallel realization of the active contour model to speed up the contour convergence process. The initially drawn contour is split into two or more independently controlled sub-contours, and each sub-contour converges independently in parallel. The final true contour is detected by combing results of all converged sub-contours. Compared with the serial realization, the contour convergence in the parallel version is more efficient, and in contrast with other parallel algorithms, the proposed method requires minimal coordination between parallel threads and thus more efficient.