Multi-scale Gaussian Segmentation via Graph Cuts

Towards the improvement of image segmentation performance, a new improved segmentation method is proposed which combines the use of multi-scale information. The multi-scale information of an image, generated by a linear Gaussian smoothing function through an interactive scheme that provides different levels of image cues which help to capture the abundant content of the image. However, Gaussian smoothing cannot generate an optimum solution to the specific scale level. According to the boundaries sensibility to scale, a convergence constraint based on the significant scale level of the segmented regions is constructed. The experimental results show that this method has superior advantages in performance and towards noise images.