2 D autoregressive model for texture analysis and synthesis

Abstract

Spatial autoregressive (AR) models have been extensively used to represent texture images in machine learning applications. This work emphasizes the contribution of 2D autoregressive models for analysis and synthesis of textural images. Autoregressive model parameters as a feature set of texture image represent texture and used for synthesis. Yule walker Least Square (LS) method has used for parameter estimation. The test statistics for choice of proper neighbourhood (N) has also been suggested. The Brodatz texture image album has chosen for the experimentation. Parameters have estimated from the textures. The test statistics decides the best neighbourhood or proper order of the model. The synthesized texture image and the original texture image have compared for perceptual similarities. It is been inferred that the proper neighbourhood for a given texture is unique and solely depends on the properties of the texture.