Rock detection via superpixel graph cuts

Xiaojin Gong,Jilin Liu

DOI: https://doi.org/10.1109/ICIP.2012.6467318

2012-01-01

ICIP

Abstract:This paper presents a rock detection method for planetary terrain scenes. Our approach first segments an image into a set of superpixels. Then we formulate the rock detection task as an energy minimization problem and solve it efficiently via a novel graph cut which is constructed on the superpixels. In order to deal with complex rock scenarios, we integrate a discriminative observation model into the graph cut framework to enhance the discrimination power. Meanwhile, a couple of features, for instance, gradient based texture and contextual shading features, are employed to characterize superpixels. With the representative features, as well as the powerful optimization model, the rock detection problem is addressed well. We test our algorithm on a real Lunar terrain image set drawn from NASA which contains diverse scenarios. The attained qualitative and quantitative results show that our algorithm is effective.