"Noise in Imaging: Focus on Correlation and Nonlinearity"

Speaker: Prof. Alessandro Foi, IEEE SPS Distinguished Lecturer School: Tampere University, Finland Understanding and characterizing noise is fundamental to the design and analysis of imaging systems and to the development of their image-processing components. This lecture examines broad classes of heteroskedastic image observations, with emphasis on noise correlation, noise anisotropy, and nonlinear effects arising from low signal-to-noise ratio (SNR), limited bit depth, and the need to maximize coverage of a narrow dynamic range. Special attention is given to the largely overlooked case of spatially correlated heteroskedastic noise subject to clipping, including extreme underexposure and overexposure. We highlight several unexpected and counterintuitive phenomena that can significantly compromise noise analysis and other stages of an image-processing pipeline unless they are properly modeled and addressed. Examples are drawn from imaging and image-processing systems used in biomedical, defense, security, and consumer applications. These include X-ray tomography, infrared thermography, video streaming, confocal fluorescence microscopy, as well as scanning electron microscopy. In addition to conventional random noise in extremely low SNR conditions, the models presented can describe various forms of structured noise and artifacts, including streak and ring artifacts in tomography, cutting artifacts introduced during volume sectioning with a diamond microtome or an ion beam, and fixed-pattern noise.