What Are Digital Images?

Digital images are discrete two-dimensional arrays of numerical values, called pixels, that encode visual information in a form suitable for storage, transmission, and computational processing. Each pixel represents the sampled intensity or color of a small area of a scene, and the collective arrangement of pixels across rows and columns reconstructs the spatial content of the original visual signal. Digital images are the foundational data type for computer vision, medical imaging, remote sensing, and visual communications, and their study encompasses the physics of image formation, the mathematics of sampling and quantization, and the algorithms used to process and analyze visual content.

The shift from analog photographic media to digital representation accelerated in the 1980s and became dominant in consumer photography with the widespread adoption of charge-coupled device (CCD) and complementary metal-oxide-semiconductor (CMOS) image sensors in the 1990s and 2000s. The IEEE Transactions on Image Processing, published since 1992, has been the primary venue for peer-reviewed research on the mathematical and computational aspects of digital imaging.

Pixel Representation and Color Models

A digital image is characterized by its spatial resolution, measured in pixels per unit length, and its radiometric resolution, the number of distinct intensity levels each pixel can take. Grayscale images assign a single intensity value, typically in an 8-bit range from 0 to 255, to each pixel. Color images represent the trichromatic response of the human visual system using multiple channels: the RGB model stores separate red, green, and blue intensity values per pixel, while the YCbCr model separates luminance from chrominance, a distinction exploited by compression algorithms and broadcast video standards. Multispectral and hyperspectral images extend beyond the visible spectrum to include near-infrared, thermal, or other sensor channels, enabling applications in agriculture, geology, and atmospheric science. Bit depth determines the dynamic range of a pixel: 8-bit encoding supports 256 levels per channel, while 16-bit or floating-point encodings preserve subtler gradations used in high dynamic range photography and scientific imaging.

Image Acquisition and Sensors

Most digital images originate from electronic sensors that convert light into electrical signals. In CMOS and CCD sensors, an array of photodiodes accumulates charge proportional to incident photon energy, and on-chip analog-to-digital converters quantize those charges into integer pixel values. Color images are typically captured by placing a Bayer filter mosaic over the sensor array, which assigns red, green, or blue filters to individual photodiodes in a repeating pattern, with twice as many green elements as red or blue to approximate the eye's higher sensitivity to mid-spectrum wavelengths. A demosaicing algorithm then interpolates full-color values at each pixel from the sparse filtered samples. Medical imaging modalities such as computed tomography (CT) and magnetic resonance imaging (MRI) produce digital images through different physical processes, with pixel values encoding X-ray attenuation or nuclear spin relaxation rather than reflected visible light. The IEEE Future Directions article on image digital sensor breakthroughs discusses recent advances in sensor architecture and in-sensor computation.

Image Processing and Compression

Raw pixel arrays are commonly processed before storage or display. Image compression reduces file size by exploiting spatial redundancy and perceptual limitations. Lossless formats such as PNG preserve pixel values exactly, while lossy formats such as JPEG apply the discrete cosine transform to blocks of pixels and discard coefficients that contribute minimally to perceived quality. The JPEG 2000 standard uses wavelet transforms for improved compression at high ratios, and the more recent AV1 and HEIC formats bring additional efficiency gains. Filtering operations, histogram equalization, edge detection, and morphological processing are applied routinely in pre-processing pipelines for machine vision, remote sensing, and medical analysis. The NASA Earthdata documentation on digital elevation and image products illustrates how image processing pipelines are applied to satellite imagery for Earth observation tasks.

Applications

Digital images have applications in a wide range of disciplines, including:

• Medical diagnosis through radiology, pathology, and surgical guidance imaging
• Remote sensing for land cover mapping, crop monitoring, and disaster assessment
• Industrial machine vision for quality control and manufacturing inspection
• Consumer photography, social media, and visual communication
• Autonomous vehicles and robotics using visual scene understanding
• Security and biometric identification through facial recognition systems