Image Segmentation

Foreground Extraction Algorithm based on an RGB-D Vision Sensor

• Main Goals

To develop an efficient foreground extraction algorithm using structure tensor and the characteristics of depth information acquired from RGB-D vision sensors such as MS Kinect.

To develop a pre-processing algorithm for interactive and energy-minimization-based segmentation approaches such as GrabCut.

• Overall Procedure

• An Real-world Example

• Application to Segmentation Techniques

Initial segmentation results obtained by applying resultant foreground and background masks to interactive segmentation (GrabCut) and matting (Spectral Matting) algorithms