Robust Edge-Based Object Segmentation by Likely Boundary Anchor Selecting and Adaptive-Thresholding Linking
Din-Yuen Chan, Wei-Cheng Wang · 2010
This paper presents a fast object contour method for segmenting the target object from a noisy image. First, a boundary-anchor identification process is employed to select the representative nodes near object’s boundary as likely boundary anchor nodes (LBANs) by performing a representative boundary-point detection/sifting (RBDS) on LBAN-based scan-lines. By connecting those LBANs, a polygonal object segmentation outcome can be yielded. Then, a piece-by-piece linking process with adaptive thresholding, named LBAN-based piecewise edge-following linking (LPEL), is to track every piece of targeted object contour LBAN-by-LBAN in exploring object’s shape. Experiment results demonstrate that the proposed method achieves more accurate object shape detection than other conventional methods in noisy images under different light projections. Such performance mainly comes from a powerful and intimate cooperation of RBDS and LPEL for the object shape detection task. Consequentially, the proposed object-segmentation algorithm is drastically fast to offer proper object segmentation in noisy images. Therefore, it could be applied into the multimedia pre-processors such as the object-based search engine for content-based image databases.