IMAGE TEXTURE SYNTHESIS-BY-ANALYSIS USING MOVING-AVERAGE MODELS

Texture synthesis is a necessary component of realistic scene generati on. In particular, it is necessary for the simulation of image backgro unds for the testing of automatic target recognizers. We present a syn thesis-by-analysis model for texture replication or simulation. This m odel can closely replicate a given textured image or produce another i mage which, although distinctly different from the original, has the s ame general visual characteristics and the same first and second-order gray-level statistics as the original image. In effect, such a synthe tic image looks like a continuation of the original scene; as if anoth er picture of the scene were taken adjacent to the original. The textu re synthesis algorithm proposed herein contains three distinct compone nts: A moving-average (MA) filter, a filter excitation function, and a gray-level histogram. The analysis portion of the texture synthesis a lgorithm derives the three from a given image. The synthesis portion c onvolves the MA filter kernel with the excitation function, adds noise , and modifies the histogram of the result. The advantages of this tex ture model over others include conceptually and computationally simple and robust parameter estimation, inherent stability, parsimony in the number of parameters, and synthesis through convolution. We 1) descri be a procedure for deriving the correct MA kernel using a signal enhan cement algorithm; 2) demonstrate the effectiveness of the model by usi ng it to mimic several diverse textured images; 3) discuss its applica bility to the problem of infrared background simulation; and 4) includ e detailed algorithms for the implementation of the model.