<title>Signal-to-noise ratio analysis of charge-coupled device imagers</title>

Abstract

We start with a physical model1 of the detector and associated electronics. Using previously derived expressions for the MTF and noise power spectrum, we extend the use of the model to include a signal-to-noise metric, the noise equivalent quanta (NEQ). This approach is then applied in a design example relevant to film and document scanning. . IntroductionCharge-coupled device (CCD) sensors are now the detector of choice for many electronic imaging applications. For integrated systems, understanding the imaging characteristics of the acquisition step is important since they limit the image information available for processing and display. System design not only involves the integration of various technologies to perform different tasks, but also the comparison of and selection between competing technical options for the same task. A consistent systematic description of signal modulation and noise degradation is valuable when identifying limitations to image quality. In particular, signal-to-noise ratio (SNR) techniques aid in the comparison of competing technologies, and in the evaluation of ultimate performance.The SNR requirements for an image acquisition subsystem are set by the intended application. Determination of these requirements involves consideration of both the input scene information and subsequent image processing and display. For the evaluation of the efficiency with which image information can be acquired the detective quantum efficiency (DQE) is often used.2 The corresponding description of the SNR, for quantum-limited applications, is the noise equivalent quantum (NEQ) exposure.3 Although these metrics were originally applied to photographic film and video cameras, the NEQ has also been used to describe the imaging characteristics of, for example, electrophotographic halftone printing,4'5 laser printers,6'7 astronomical detectors8 and medical diagnostic systems.9'1° The noise equivalent input approach has also been applied to the performance of human 1 The NEQ describes the absolute SNR in terms of an equivalent input quantum exposure. It can, therefore, be used to compare imaging systems and components that use different technologies and physical mechanisms. Here we only address CCD imagers with the goal of expressing the NEQ in terms of the various model design parameters.2 . CCD Imager Model Figure 1 shows a functional block diagram of a CCD sensor in the focal plane of an optical system and its associated electronics. Incident photons are detected in the photosensitive layer simultaneously at each photosite (pixel) location across the detector. The energy released by the absorbed photons generates electron-hole pairs. The electrons then migrate to potential wells in the device so that the collected charge represents the signal integrated over each photosite. This charge, collected during a fixed time interval, is then transferred from the potential wells through a shift register to the readout node. The final step is the readout of the charge packets, analog amplification and quantization of...

Scene-and-Process-Dependent Spatial Image Quality Metrics

“…The STV-IQMs and MF-IQMs -referred to in this paper as engineering metrics -employ standard spatial system performance measures such as the Modulation Transfer Function (MTF) and Noise Power Spectrum (NPS), and threshold contrast sensitivity functions (CSF) describing visual spatial sensitivity. The Noise Equivalent Quanta (NEQ) signal-to-noise measure is core to the most relevant STV-IQMs and is applied widely in capture system and sensor modelling [6]- [8]; it also uses the MTF and NPS.…”

Local Image Reconstruction and Subpixel Restoration Algorithms

“…Since all physically realizable imaging systems must have a PSF of finite spatial extent, point sampling, the basis for most reconstruction algorithms, is not physically realizable. Researchers have developed complex models of CCD blurring in terms of MTFs (e.g., [20], [3]). Unfortunately, dealing with multi-pixel blur increases the computational complexity because the algorithm will no longer be local.…”