Various embodiments are disclosed for generating depth maps. One embodiment is a method implemented in an image processing device. The method comprises retrieving, by the image processing device, a (2D) image; and determining, by the image processing device, at least one region within the (2D) image having a high gradient characteristic relative to other regions within the (2D) image. The method further comprises identifying, by the image processing device, an out-of-focus region based on the at least one region having a high gradient characteristic; and deriving, by the image processing device, a color model according to the out-of-focus region. Based on the color model, the image processing device provides a depth map for (2D)-to-stereoscopic conversion.