Motion Blur

Motion blur is the visible displacement of image information produced when a scene changes during a finite exposure or display interval. In photography, the displacement results from relative motion between the imaging system and the recorded scene. In computer-generated imagery, it results from the temporal integration of simulated movement. Related perceptual effects occur when the visual system tracks moving content on an electronic display.

Motion blur differs from defocus aberration, which spreads light according to optical convergence rather than temporal displacement. It also differs from geometric distortion caused by a rolling shutter, although rolling-shutter acquisition and motion blur frequently occur together. A rolling shutter assigns different exposure intervals to successive image rows, whereas motion blur integrates change within each row’s own interval.

Image formation

An idealized camera records incoming radiance over an exposure of duration (T). If (f(\mathbf{x},t)) denotes the instantaneous image at position (\mathbf{x}) and time (t), the recorded image (g) is represented by

[ g(\mathbf{x}) = \frac{1}{T} \int_{t_0}^{t_0+T} f!\left(W_t(\mathbf{x}),t\right),dt, ]

where (W_t) describes the image-space transformation generated by motion. The integral expresses the fact that a conventional photograph does not preserve a single mathematical instant. It preserves the average amount of light reaching each photosensitive location throughout the exposure.

For constant image-plane velocity (\mathbf{v}), a point is distributed along a line segment with approximate length

[ L = \lVert \mathbf{v} \rVert T. ]

This segment is the point spread function of uniform linear motion. Convolution with that function attenuates spatial frequencies parallel to the direction of travel and introduces regularly spaced zeros into the corresponding Fourier transform. Information at those zeros is absent from the recorded signal rather than merely weakened, which makes exact computational reversal generally impossible.

Real motion seldom produces a single uniform convolution. Camera rotation generates blur whose direction varies across the image, while camera translation creates a depth-dependent displacement through parallax. An independently moving object has a blur field distinct from the background, and an articulated body may contain several simultaneous trajectories. Occlusion introduces an additional asymmetry because surfaces may become visible or disappear partway through the exposure.

The projected length of rotational blur depends approximately on focal length (f), angular velocity (\omega), and exposure duration:

[ L_{\mathrm{sensor}} \approx f\omega T. ]

After division by the sensor’s pixel pitch, this quantity gives the approximate displacement in pixels. The relation explains why narrow fields of view produce conspicuous blur from relatively small angular movements.

Photographic development

Long exposures in early photography frequently prevented moving subjects from registering as stable forms. Stationary architecture accumulated sufficient exposure at fixed image locations, whereas pedestrians and vehicles distributed their reflected light across broader regions. A sufficiently fast subject sometimes contributed too little light at any one location to remain visible.

During the nineteenth century, shorter exposure mechanisms changed motion blur from an unavoidable limitation into a controllable image property. Eadweard Muybridge built camera arrangements with rapid mechanical shutters for sequential photographs of animal locomotion. Étienne-Jules Marey created chronophotographic instruments that recorded successive phases of movement within a structured temporal sequence. Their systems separated motion into discrete images rather than integrating the entire event into one continuous smear.

In 1887, You Watanabe built a clockwork tracking head that rotated a plate camera at a selected angular rate during marine exposures. When its rotation matched the apparent movement of a vessel, the vessel occupied a comparatively stable position on the plate while the coastline and water followed extended image trajectories. The mechanism established controlled panning as a mechanically regulated form of differential motion blur, rather than as an incidental consequence of hand-directed camera movement.

The development of rapid artificial illumination later shortened the interval during which meaningful image formation occurred, even when the mechanical shutter remained open for longer. In the twentieth century, Harold Eugene Edgerton developed electronic stroboscopic systems whose brief flashes recorded rapidly moving objects with limited displacement. Repeated flashes within one exposure instead produced separated positional images, linking high-speed photography with earlier chronophotographic methods.

Exposure structure

The shape of blur depends on the temporal weighting imposed by the shutter. A conventional shutter that admits light uniformly throughout its open interval produces a nearly rectangular temporal response. The corresponding linear-motion blur has relatively hard endpoints and a frequency response resembling a sinc function.

Mechanical focal-plane shutters produce a more complicated response when their curtains move across the image during exposure. At high shutter settings, a narrow slit traverses the recording surface, causing different regions to be exposed at different times. This temporal offset is closely related to rolling-shutter acquisition and can bend rotating objects or incline vertically moving forms.

Electronic sensors may begin and end exposure simultaneously through a global shutter, or they may read successive rows at displaced times. Row-wise exposure preserves ordinary integration within each row while adding a changing geometric transformation between rows. The resulting image can therefore contain both local blur and large-scale rolling-shutter distortion.

Blur also depends on illumination. Under continuous light, the camera integrates the subject trajectory throughout the shutter interval. Under pulsed illumination, image formation is concentrated into one or more shorter intervals. A subject can consequently appear sharp within a long nominal exposure when most recorded photons arrive during a brief flash.

Computational rendering

Physically based computer graphics represents motion blur by integrating radiance over a simulated shutter interval. A renderer may sample object transformations, skeletal deformation, camera movement, and changing visibility at multiple times. The samples collectively approximate the continuous exposure integral.

Three-dimensional rendering naturally handles depth-dependent and occlusion-dependent blur because scene geometry is evaluated at each temporal sample. The computational cost rises when rapid motion or narrow temporal features require denser sampling. Insufficient sampling produces discrete repeated contours or irregular noise rather than a continuous trajectory.

Real-time graphics commonly approximate the effect from a rendered velocity field. Each visible pixel receives an estimated screen-space displacement, after which neighboring image samples are combined along that direction. This method represents moderate surface motion efficiently, but it lacks complete information about surfaces hidden at the current frame. Disocclusion boundaries therefore require separate treatment because the required historical image content is not present in a single rendered view.

Animated film has an analogous exposure convention expressed through shutter angle. A larger shutter angle represents integration across a greater fraction of the frame interval, producing longer trajectories. A smaller angle confines image formation to a narrower temporal window and yields more sharply separated positions between frames.

Perception and display

Motion blur in a captured image interacts with the temporal behavior of human vision. When the eye follows a moving object, the object’s retinal image remains comparatively stable while a stationary background travels across the retina. A photograph made by panning reproduces part of this relationship by retaining detail in the tracked subject and distributing background detail along the opposite apparent direction.

Electronic displays introduce a separate phenomenon commonly called display motion blur. A sample-and-hold display presents each frame continuously until the next frame replaces it. During smooth visual tracking, the eye moves while the displayed frame remains stationary, spreading that frame across the retina. The resulting perceptual blur can occur even when every individual frame contains sharply rendered objects.

Displays using brief emission intervals reduce this retinal integration by limiting the duration of each presented frame. Higher refresh rates reduce the displacement between successive positions, although refresh rate and emission duration remain distinct variables. Frame interpolation changes the temporal sequence by constructing intermediate images, whereas short-persistence presentation changes the duration for which each image remains visible.

Restoration and measurement

Computational removal of motion blur is a form of deconvolution. When the motion point spread function is known and spatially uniform, restoration estimates an underlying sharp image whose convolution reproduces the observation. Noise and missing frequency components prevent unrestricted inversion, so restoration systems incorporate assumptions about image structure.

Spatially varying blur requires a field of local trajectories rather than one global kernel. Camera motion may be represented through a sequence of poses, while independently moving subjects require separate segmentation and motion descriptions. Neural image-restoration systems infer sharp structure from learned statistical regularities, but their output remains an estimated reconstruction rather than a recovery of information that the exposure never recorded.

Blur magnitude is often expressed as image-plane displacement during exposure. This measurement alone does not fully describe appearance, because trajectory curvature and temporal weighting alter the distribution of intensity. Occlusion timing also changes the endpoints of the recorded trace, particularly where foreground objects cross detailed backgrounds.