Reconstruction of the Degraded Image: Restoration

Abstract

In this chapter, a set of techniques that perform quantitative corrections on the image to compensate for the degradations introduced during the acquisition and transmission process are described. These degradations are represented by the fog or blurring effect caused by the optical system and by the motion of the object or the observer, by the noise caused by the opto-electronic system, and by the nonlinear response of the sensors. It is also introduced by random noise due to atmospheric turbulence or, more generally, from the process of digitization and transmission. While the enhancement techniques tend to reduce the degradations in the image in qualitative terms, improving their visual quality even when there is no knowledge of the degradation model is fundamental. To this aim, restoration techniques are used to eliminate or quantitatively attenuate the degradations present in the image, starting also from the hypothesis of knowledge of degradation models. Restoration techniques essentially recover the original image (without degradation) from the degraded image through an inverse process of the hypothesized degradation model (for example, de-blurring, additive noise, etc). The degraded image is considered modeled by convolution of the original image, the degradation function, and additive noise. The process of restoration can be modeled by a deconvolution process of this degraded image to obtain noiselessly and deblurred original image. A variety of noise models and restoration methods are described such as Weiner filter, inverse filter, constrained least-square filter, blind deconvolution, spatial and frequency filer, adaptive filter, and optimal filter. Some of these methods are either linear or nonlinear and a comparison of restoration approaches are evaluated.