Class-Specified segmentation with multi-scale superpixels

摘要: This paper proposes a class-specified segmentation method, which can not only segment foreground objects from background at pixel level, but also parse images. Such is very helpful to many other computer vision tasks including computational photography. The novelty of our method that we use multi-scale superpixels effectively extract object-level regions instead using single scale superpixels. contextual information across scales and the spatial coherency neighboring in same are represented integrated via Conditional Random Field model on Compared with methods have ever used superpixel extraction together across-scale modeling, has fewer free parameters simpler effective. superiority compared related approaches, demonstrated two widely datasets Graz02 MSRC.