A novel dense stereo matching algorithm based on propagation within a Voronoi decomposition of the image plane is proposed. A weighted sum of squared differences is developed as the cost function. The size of the window is adaptive; it is set in inverse proportion to the texture density inside it for reliable propagation. A simple way to detect possible areas of depth discontinuities is developed to determine the search regions. A significant merit of the algorithm is that it can be applied to a wide variety of images, including those with large disparities, with or without rectification. It has been verified on real images, with satisfactory results.