A comparison of genetic algorithm, regression, and Newton's method for parameter estimation of texture models

The estimation of parameter values in stochastic models for texture image data is the focus of this paper. We present a comparison of three methods that seek optimal solutions to a problem of stochastic model selection in texture data. The problem is to find parameter values for the model that "best" fits the texture data. The three methods used to solve this problem are genetic algorithms, logistic regression, and Newton's method. We present a comparison of the results of these three techniques to both synthetic and real data. For synthetic data, the values estimated by the genetic algorithm, logistic regression, and Newton's method gave textures that were practically indistinguishable from the original textures. For real data, the results are encouraging and warrant continued investigation.