Hyperspectral imaging is an emerging non-invasive modality that has shown great potential in numerous biomedical applications such as cancer diagnosis, burn depth assessment, and early caries detection, as well as in other fields including pharmacy, food industry, agriculture, remote sensing and astronomy. Hyperspectral imaging systems produce a stack of images acquired at many different wavelengths that provide information about the spectral content of the object and its spatial distribution. Acquisition of hyperspectral images typically involves illumination of the observed object by a broadband light source. The diffused or transmitted light is collected by the front lens and directed onto a dispersive element from where it is refocused onto the detector array. Properties of the dispersive element, front lens and misalignments of the optical elements contribute to positionally variant displacements and blur that can significantly degrade the overall quality of the acquired images. In general, hyperspectral images can be acquired in four different ways: whiskbroom, pushbroom, staring and snapshot configuration. In the case of whiskbroom and pushbroom systems, hyperspectral image is formed by spatially scanning the object in each pixel or line of pixels, respectively. Staring systems conduct a spectral scan of the object, acquiring 2D images at the selected spectral bands. On the other hand, snapshot systems allow simultaneous acquisition of the spatial and spectral information.
The image formation process can be mathematically formulated by convolution of the observed scene with the response function, that models the aberrations introduced by the hyperspectral imaging system. Having an accurate estimate of the response function, deconvolution can invert the image formation process, obtaining an undistorted high-resolution estimate of the observed scene. From the theory of linear systems it is well known that the system can be fully characterized if the response to a standard test function is known in each state of the system. The main goal of this thesis is to provide the users of hyperspectral imaging systems with novel methods and tools to accurately measure and identify the response function that can be employed in subsequent deconvolution-based image restoration, reducing the effects of displacements and blur in the acquired images.
The importance of hyperspectral imaging system characterization is recognized in the field of remote sensing where several studies have been published in recent years. However, it is overlooked that image deconvolution could be used to simultaneously reduce the effect of displacements and blur arising from the optical system. Furthermore, the proposed characterization methods require the lens working distance to be set to infinity, which is rarely the case in laboratory applications of hyperspectral imaging systems. Finally, in most of the laboratory hyperspectral imaging systems merely a flat-field correction is applied to eliminate the effect of illumination non-uniformity and sensor sensitivity, fully neglecting the effects of system optics on the acquired images.
In the first part of this thesis, we devise a complete restoration procedure for pushbroom hyperspectral imaging systems, refining the previous work on the characterization of laboratory pushbroom hyperspectral imaging systems in a way that allows efficient deconvolution based image restoration. In the second part of the thesis, we propose and analyze a novel calibration target that offers a simple solution for direct and highly accurate 3D response function measurements of diffraction limited hyperspectral imaging systems.