Synthesis of 2D and 3D images by generalized long-correlation models

The authors consider the modeling of two-dimensional (2D) and 3D natural scenes by using a single statistical model. A model developed can synthesize both 2D and 3D images that resemble real textures and natural terrains. The new model is a generalization of long-correlation models. The relationship between the new model and fractals is discussed. The performance of generalized long-correlation models is demonstrated by synthesizing various 2D and 3D images resembling real 3D terrains and 2D textures. The authors also present an algorithm for estimating the parameters of generalized long-correlation models. The images generated by estimated parameters look similar to the original images