Abstract
Imaging with object-specific feature measurements has been of interest due to its wide variety of applications in signal and image representation, enhancement, classification and compression. Research has taken place on the performance of linear grayscale face features with respect to reconstruction MSE and feature MSE in different noise environments [1]. In this paper, we analyze a novel multispectral imaging system that measures linear features of an object directly. The system is designed to be energy efficient when compared to a conventional imager. A small number of direct feature measurements exploit the Fellgett advantage to achieve higher fidelity than features that are post-processed from images generated by conventional imaging. A dual-rail optical system with optimized irradiance sharing is designed for this purpose. The system performance is quantified using Karhunen-Loeve features in terms of feature RMSE and reconstruction RMSE, in the presence of noise. Performance comparison for varying block-size, number of features etc. is performed for RGB face images and Landsat-7 images, with their grayscale counterparts [2]. Results show that in high noise situations a multiplex imager provides images with higher fidelity than a conventional imager does.
© 2003 Optical Society of America
PDF ArticleMore Like This
Wenli Huang, Barry L. Shoop, Eugene K. Ressler, Cedric Bonamigo, and Alexandre Garcia
JWA55 Frontiers in Optics (FiO) 2005
Volker Seyfried, Holger Birk, Rafael Storz, and Heinrich Ulrich
5139_147 European Conference on Biomedical Optics (ECBO) 2003
V.V. Shapovalov, B.S. Gurevich, S.V. Andreyev, A.V. Belyaev, and V.N. Chelak
808724 European Conference on Biomedical Optics (ECBO) 2011