Image based road detection is a vital and challenging task in many real-world applications such as autonomous driving, road departure warning. In this paper, we study the problem of unstructured road detection from a color image. The algorithm we propose decomposes the road detection into three steps: the estimation of the horizon line by similarity measurement, followed by the road area prediction based upon the estimated horizon line, and region growing to segment the image into road and non-road regions. The road area prediction provides good road representations. Then region growing is effective to detect the road region utilizing the feature extracted from the road area. Experimental results demonstrate that the proposed algorithm is accurate and efficient.