TABLE OF CONTENTS Image processors could be categorized into different levels by the human vision standard. Lower-level ones are to clean and enhance observations, interpolate missing image data, or identify regions occupied by objects without telling what they are. For example, counting the number of cells in a blood-sample image falls into this category. Higher-level processors are to recognize object features and identify the associated hidden real-world contexts, such as face recognition for video surveillance and terrain reading for automatic piloting.
In this sense, the human vision system, involving both optical and neuronal activities, is a highly advanced and complex system of image processors. The present book focuses only on lower-level vision activities and their associated image processors such as denoising, deblurring, inpainting and interpolation, and segmentation. Not only do they frequently emerge in a variety of scientific or industrial applications, but also there have been accumulated numerous significant contributions by applied mathematicians and imaging scientists.
In this chapter, we illustrate how to develop efficient image denoising schemes based upon appropriate image models or representations discussed in the previous chapters.
Noise is ubiquitous and noisy but not always annoying.
From the standpoint of statistical mechanics, noise is intrinsic to any multibody system. For image and vision analysis, such a system can be an electric current of many electrons flowing through a resistor or a circuit (in electrical imaging devices), a beam of photons striking the layers of the two retinas,...
