FAST ROBUST OBJECT SEGMENTATION BY PROGRESSIVE SHAPE-ANCHOR SELECTING AND ADAPTIVE-THRESHOLDING PIECEWISE LINKING

摘要

This paper presents a fast robust object segmentation method for segmenting the target object from a noisy image. First, an anchor-node identification process is employed to select the representative nodes on object's boundaries as shaping anchor nodes (SANs) by performing a boundary-point detecting-sifting (BPDS) on SAN-based scan-lines. Then, an adaptive-thresholding piecewise linking process, named SAN-based piecewise shape linking (SPSL), is used to render a closed contour passing every SAN for getting a precise profile for the image object targeted. When linking of the closed contour fails for some SANs near weak-edge object boundary, the resultant disconnections can be recovered by simply iterative refinement linking. Experiment results demonstrate that the proposed method achieves more accurate object shape detection than other conventional methods in noisy images under different light projections, while the required process cost is not complicated. Such performance mainly comes from the tight cooperation of BPDS and SPSL for the object shape detection task. Consequentially, the proposed object-segmentation method is drastically fast to offer proper object segmentation in noisy images.