Salient Boundary Detection in the Gradient Spaces

In shape learning for a digital image, it is a common that one would encountered two major problems, which are noise and missing edges. The existing research approaches aim to reduce noise, while at the same time to recover the missing edges. However, noise reductions using contour detector or smoothening techniques could indirectly distort the salient boundary. Most of the research works only focus on the preservation of blurred and small edges, but failed in containing the shape of objects in the images. In this research, a comprehensive study based on four gradient fields is conducted. In this study, salient boundary detection (SBD) method is proposed. This method not only eliminates noise, but also recovers the missing edges, that forms the shape outline for objects in the images. As a result, an average probability rand index (PRI) has achieved 0.8031, tested on 100 ground truth images in the Berkeley Segmentation Database (BSD500).