Abstract. The improvement of image quality in the infrared range on silicon-based sensors is one major topics using these long wavelength channels for geometric measurements. The reason behind the bad quality of infrared images in comparison with the visible sampling range is explained by the wavelength response dependency of the silicon. Photons are able to either to pass the sensitive range of the pixel or can tunnel to the neighbor pixel. This effect leads to blurred images, which will not only increase the uncertainty of measurement, but also the aesthetical quality of the image. In this paper, methods to improve the image quality using blind convolution as well as a special infrared focusing to improve the sharpness will be presented.
IntroductionMultispectral and hyperspectral imaging technologies allow detecting several features in industrial inspections tasks using the different spectral channels [1] its applications range from cultural heritage, arts, remote sensing, environmental monitoring, medicine, biology, food quality control, military applications, factory automation and manufacturing [2]- [7]. Besides an accurate registration of images acquired at different wavelengths for the spatial-spectral utilization of multispectral data [8], the consideration of the semiconductor characteristics is a big issue. Therefore, in [9] a method for the characterization of pixel inhomogeneity's caused by the manufacturing process and read out process is proposed. These investigations show how the specific data can be processed to compensate the inhomogeneity's as function of the current integration time. A fact which is actually not discussed in these works, is the crosstalk characteristics between the pixels in dependence of the wavelength. The material characteristic plays an important role. The characteristics of penetration depth versus absorption of photons for different materials are given in [10]. It is shown in [10] and [11] that in the near infrared region (NIR) the generation of electrons will happen in an range of tens to hundreds of microns inside the semiconductor. Therewith it is possible that the electrons flow into neighbor pixels, which is presented in [12]. Finally, these effects in the modular transfer function of the image sensor and the images have a blurred appearance as well as blurred information across edges, which account for the deterioration of the accuracy of geometrical measurements in the NIR regime: blurred images have less sharp edges (see Figure 2) and edge detection based algorithms rely on precise estimation of edge's position. Unfortunately, this effect has a statistical behavior so a discrete model based deconvolution of the blurred image is not possible. In this paper, an approach to improve this characteristic using a statistical method based on a combined image restauration algorithm is part of the discussion. Other sources of instabilities and how to tackle them are discussed in [1], [8]-[9], [13].
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.