Some scenarios require performance estimation of an imaging or a computer vision system prior to its actualoperation such as in system design, as well as in tasks of high risk or cost. To predict the performance, we propose an image-based approach that accounts for underlying image-formation processes while using real image data. We give a detailed description of image formation from scene photons to image gray levels. This analysis includes all the optical, electrical, and digital sources of signal distortion and noise. On the basis of this analysis and our access to the camera parameters, we devise a simple image-based algorithm. It transforms a baseline high-quality image to render an estimated outcome of the system we wish to operate or design. We demonstrate our approach on thermal imaging systems (infrared spectrum, 3-5 (mu)m).
展开▼
