What Is Multidimensional Signal Processing?

Multidimensional signal processing is a branch of signal processing concerned with signals that vary as functions of two or more independent variables, such as spatial coordinates, time and space together, or multiple frequency dimensions. While classical one-dimensional (1D) signal processing operates on sequences indexed by a single variable such as time, multidimensional (nD) processing generalizes the foundational operations of filtering, spectral analysis, and sampling to signals defined over 2D, 3D, or higher-dimensional index sets. The field encompasses the mathematical theory needed to extend 1D results to higher dimensions and the algorithms that make this theory computationally feasible for real data sets.

Multidimensional signal processing draws from functional analysis, linear algebra, and discrete mathematics, and it intersects with image processing, geophysics, medical imaging, and computer vision. The critical distinction from 1D processing is that many of the simplifications that make 1D filter design tractable, including the close connection between linear-phase filters and symmetric impulse responses and the factorizability of spectrum into frequency and time, do not generalize straightforwardly to higher dimensions, requiring new design methods and different intuitions.

Multidimensional Filter Design

Designing filters for nD signals involves specifying a desired frequency response over a multidimensional frequency space and constructing a corresponding filter with finite or infinite impulse response. In 2D, the frequency response is a surface over the (omega-x, omega-y) plane, and ideal lowpass filters have circular or elliptical passbands rather than the simple band-limited intervals of 1D filters. Frequency-domain specifications are typically approximated using window-based methods, transformation from 1D designs using the McClellan transformation, or iterative optimization. For separable filters, the 2D problem factors into two 1D designs applied along each dimension independently, but non-separable designs require joint optimization across both dimensions. The IEEE Signal Processing Society's Image, Video, and Multidimensional Signal Processing Technical Committee coordinates research and standards activity in this area and organizes the biennial IVMSP workshop.

Spectral Analysis and Transforms

The Discrete Fourier Transform (DFT) extends naturally to multiple dimensions by applying the transform sequentially along each index dimension, producing a multidimensional spectrum where each frequency component corresponds to a spatial or spatio-temporal oscillation pattern. The 2D DFT and its fast algorithm, the 2D FFT, are foundational to image filtering in the frequency domain. Non-separable transforms, including the discrete cosine transform (DCT) applied in the JPEG and HEVC video standards, and wavelet transforms with 2D filter banks, provide alternative spectral decompositions suited to directional features and multiresolution analysis. Textbook treatment of these transforms and the sampling theory that governs their application to continuous-domain signals is found in Multidimensional Signal, Image, and Video Processing and Coding by John W. Woods, which covers both the theoretical framework and practical algorithms for 2D and 3D cases including super-resolution and non-local processing methods.

Image and Video Processing

Image processing is the largest and most developed application domain of multidimensional signal processing, treating a digital image as a 2D signal and a video sequence as a 3D signal indexed by horizontal position, vertical position, and time. Operations such as spatial filtering, edge detection, noise reduction, and compression are implemented using the nD filter and transform methods developed in the field. The IEEE IVMSP workshop series, with the 15th edition planned for 2026 in Shenzhen, China, covers topics ranging from super-resolution and inverse problems to learned image compression and computational imaging, reflecting the growing integration of multidimensional signal processing methods with machine learning.

Applications

Multidimensional signal processing has applications across a wide range of scientific and engineering disciplines, including:

• Medical imaging reconstruction in MRI, CT, and ultrasound systems
• Remote sensing and synthetic aperture radar image formation
• Video coding and compression standards including HEVC and AV1
• Seismic data processing for subsurface geological imaging
• Computational microscopy and fluorescence imaging in biological research