On-board Image Quality Assessment for a Satellite

An image captured by a sensor is not an exact reproduction of the scene viewed. If a point of light is viewed, the optical system will blur or spread the light to some degree, characterised by the point spread function (PSF). In the mathematical model of the imaging system, the PSF is a weighting function for spatial convolution [94, pp. 7891]. It can be considered as the spatial responsiveness of the imaging system. If the imaging system becomes defocused, the PSF shape will be aected and the image will become blurry. Estimation of image sharpness, focus or blur spans a variety of elds. Many dierent techniques can be employed depending on the application. Measurement of image sharpness is a crucial part of edge detection algorithms [47, pp. 572585]. These algorithms make no assumptions about the image acquisition chain; they simply evaluate the sharpness of edges in an image. Estimation of optimal focus is an important problem with practical application in autofocus algorithms for cameras [121, 115]. These algorithms rely on a sequence of images from the same subject. Typically an image sharpness measure is computed for each image in the sequence. The measure is compared across all the images in the sequence. The sharpest image is assumed to correspond to an in-focus imaging system. More advanced methods use focus measures that are invariant to illumination changes [115]. Others estimate the PSF of the imaging system so that the focus measure is invariant of the imaged object [121]. Estimation of PSF is also required during restoration of blurred images (called deconvolution). Sometimes it is possible to estimate the PSF by using 88 4.1 Introduction 89 knowledge of the imaged object [47, pp. 256257] [17]. When the imaged object is unknown, blind estimation of PSF is used. These techniques use only a single potentially degraded image to estimate PSF and are discussed further in section 4.2.1. In the remainder of the introduction the degraded image model is explained (section 4.1.1) and defocus estimation in the context of image quality assessment is discussed (section 4.1.2). An overview of existing literature is presented in section 4.2: section 4.2.1 introduces the eld of blur estimation; section 4.2.2 discusses the theoretical base of the class of blur estimation algorithms used; section 4.2.3 explains some of the shortcomings of the methods in existing literature. Angular spectral smoothing, a new technique based on the method discussed in 4.2.2, …