Different labels in energy minimized via graph cuts for stereo matching

Several new algorithms for stereo matching based on graph cuts have recently been developed for pixel-labeling tasks such as depth. Such problem can be expressed as Markov random fields, yet the resulting energy minimization problems have been widely viewed as intractable. In this paper, we describe energy minimization based on graph cuts. The data term of energy function adopts three different approaches to obtain labels such as Birchfield and Tomasi [2] approach, the absolute difference between corresponding pixels and the squared difference between corresponding pixels. The smoothness term of energy function uses Expansion moves and Swap moves to compute a local minimum. The experimental results demonstrate differences between the three approaches efficiently. We compared the solutions quality and running time and analyzed excellence and disadvantage.