ROBUST IMAGE MODELS WITH APPLICATION
Abstract
Robust parameter estimations in two different types of image models, causal autoregressive models and the long correlation models, are considered. The robust methods in image models are also applied to some important image processing problems such as image segmentation by texture property and image restoration in the presence of impulse noise. A convenient estimation algorithm is developed for long correlation models. This algorithm is computationally more attractive than the maximum likelihood estimator. The new estimator has low mean square error (close to Cramer-Rao lower bound) over a large class of noise distributions. A texture boundary detection algorithm based on a long correlation model is also developed and tested with real images. This algorithm does not require any prior knowledge of the number of regions or types of textures. The texture boundaries cannot effectively be detected by traditional edge detection algorithms or pattern classification methods. Experimental results shows that the algorithm based on a long correlation model successfully detects texture boundaries on real images when other traditional edge detection algorithms fail. Robust estimation algorithms for two different outlier processes in causal autoregressive models are developed. These algorithms are based on robust M-estimators. Theoretical properties of the robust estimation algorithms are presented. The robustness of the estimators are also shown in the experiment. The robust estimation algorithm for causal autoregressive models is applied to the image restoration problem. Traditionally, median or (alpha)-trimmed mean filters are used, but these methods result in blurred images. The restoration method based on robust image model cleans out impulse noise without involving any blurring of the image. Experimental results show that the quality of images restored by the model-based method is much superior to the images restored by other traditional methods.
Degree
Ph.D.
Subject Area
Electrical engineering
Off-Campus Purdue Users:
To access this dissertation, please log in to our
proxy server.