This paper presents a new method to extract foreground using Graph-cut on color and texture combination. Traditional graph-cut algorithm only used intensity at pixel level, this amount of information is insufficient for complex situations in which objects are camouflaged or in uneven exposure condition. Our method makes use of texture additionally from dense SIFT sampling to segment robustly hard images, Graph-cut algorithm then fuses color and texture to build graph for optimization; and vector quantization will be also applied to avoid over-fitting problem and improve performance. Experiments on Mircrosoft Grabcut and hard cases in Berkeley and ETHZ datasets using our method show comparable segmentation results against the state-of-the-art method, and outperform in challenging images.