What Is Image Sampling?

Image sampling is the process of converting a continuous spatial signal (the light intensity across a scene) into a discrete array of measurements by recording values at a finite set of spatial locations. In a digital camera, sampling occurs at the sensor plane, where a rectangular grid of photodetectors records one (or several) intensity values per grid cell, producing the pixel array that constitutes the digital image. The theory of image sampling sits within the broader framework of the Nyquist-Shannon sampling theorem and governs the fidelity with which the discretized image can represent the original continuous scene. The field draws on signal processing, optical physics, and information theory.

Sampling is never lossless in practice. The moment a continuous image is discretized onto a finite grid, spatial frequencies above half the sampling rate (the Nyquist frequency) can no longer be faithfully represented. If those frequencies are present in the image, they create aliasing: high-frequency content folds back into the low-frequency range and appears as spurious patterns such as Moire fringes, jagged diagonal edges, or false color. Managing aliasing is one of the central design problems of any digital imaging system.

Nyquist Sampling and Aliasing

The Nyquist-Shannon sampling theorem states that a signal band-limited to a maximum spatial frequency f_max can be perfectly reconstructed from samples taken at a rate of at least 2*f_max (the Nyquist rate). Applied to imaging, this means the pixel pitch of a sensor must be small enough to sample at twice the highest spatial frequency present in the optical image formed on the sensor. Image sampling and aliasing, covered in the MIT Foundations of Computer Vision textbook, explains how the modulation transfer function of the optics sets the effective bandwidth of the incoming image, and how the relationship between optical bandwidth and sensor pitch determines whether aliasing will occur.

Anti-aliasing filters (optical low-pass filters, OLPF) are placed in front of the sensor in many digital cameras to attenuate spatial frequencies approaching the Nyquist limit before they are sampled. These are commonly implemented as birefringent crystal plates that slightly blur the image by splitting each point into multiple shifted copies. Dimensioning of birefringent anti-aliasing filters for digital cameras, published in an IEEE conference, treats the design trade-off between aliasing suppression and the sharpness penalty imposed by the blur. Some cameras omit the OLPF entirely to maximize perceived sharpness, relying on demosaicing and post-processing to manage residual aliasing.

Non-Uniform and Compressive Sampling

Standard digital sensors sample on a regular rectangular grid, but the sampling locations need not be uniformly spaced. Non-uniform sampling places samples at irregular positions, potentially allocating higher density to regions with rapid spatial variation and lower density to smooth regions. Reconstruction from non-uniform samples requires specialized interpolation methods based on iterative algorithms or frame theory. Compressed sensing generalizes this idea further: if the image is known to be sparse in some transform domain (wavelet coefficients, for example), far fewer random or structured measurements than the Nyquist rate would demand are sufficient for exact or approximate recovery. Dynamic non-regular sampling using frequency-selective reconstruction, published in IEEE Transactions on Image Processing, demonstrates an adaptive sensor design that selects sampling locations based on local signal content to improve effective resolution within a fixed measurement budget.

Applications

Image sampling has applications in a wide range of fields, including:

• Digital camera and scanner sensor design and optical system specification
• Medical imaging devices such as CT, MRI, and ultrasound scanners
• Remote sensing and satellite imaging systems with fixed detector arrays
• Video acquisition and display at standard and high frame rates
• Compressive sensing-based single-pixel and coded aperture cameras