Why a point can become a disc
A pinhole selects essentially one ray from each scene point. A lens with a finite aperture collects a bundle: more light can reach the sensor, but the rays from a point meet at only one image distance. If the sensor is elsewhere, the bundle cuts it in a finite patch. That patch is defocus blur, not necessarily a faulty calibration or motion. To quantify it we need a model of the lens, the aperture, and the sensor position, not just a projective point-to-point map.
Definition 1.21.1
(Thin-lens camera model)
We use an ideal thin converging lens in air, focal length , at . A scene point is at distance on one side; a planar sensor is at distance on the other. Distances are positive and use the same units. The lens is focused at a chosen distance , so . The lens obeys the paraxial thin-lens equation
Here is the real-image distance of the particular point, not necessarily the fixed sensor distance . We assume a circular, centered aperture of diameter , no occlusion or aberrations, and geometrical rays in the paraxial approximation. The ideal lens maps the ray through its center straight on and bends rays from the same point to meet at the image point. This is an optical model and lens equation assumed here; the following similar-triangle results are consequences, not a derivation of real lens design.
Deriving the circle of confusion
Theorem 1.21.2
(Geometric defocus diameter)
Under these assumptions the rays through the aperture from one point form, at the sensor, a disc of diameter
It collapses to a point at . The formula describes the support of the ideal geometric blur, not its brightness profile or a diffraction-limited point-spread function.
Proof
(Similar triangles at the sensor)
At the lens plane the extreme rays are separated by ; at their common image point a distance away they have separation zero. Linear ray propagation makes their separation at equal to . The lens-plane aperture maps to the sensor plane by a uniform signed scaling about the chief ray, followed by a translation. Its circular image is therefore the stated disc; for an off-axis point its center is displaced, but the diameter is the same in this ideal model. From the lens equation, and , so
Taking absolute values gives the second expression. At all rays meet at the sensor. The limit is allowed only as a limit: its image distance tends to . The proof assumes the complete aperture reaches the sensor; clipping changes the shape and support.
Proof X-ray
Which distance is fixed?
Changing the object distance changes , while the already positioned sensor remains at . Substituting for the sensor position at every depth would refocus the camera for every point and incorrectly predict zero blur everywhere. The absolute value is essential because the sensor can lie on either side of the ray intersection.
A numerical focus comparison
Worked example 1.21.3
(One camera focused at one meter)
Use millimeters: , and , corresponding to an ideal f-number . The fixed sensor distance is
For a point at mm, its rays meet at mm. At the fixed sensor their support diameter is
At the focus distance the same calculation gives . If the sensor has mm pixel pitch, this ideal disc spans roughly pitches; that ratio alone does not specify its sampled pixel values, which depend on the intensity distribution and pixel integration. The estimate is deliberately large enough to make the geometric effect visible; aberration and diffraction are omitted.
An acceptable blur criterion gives a depth interval
Depth of field is not a second plane on which all points are exactly sharp. It is defined relative to a chosen blur tolerance.
Definition 1.21.4
(Acceptable blur and geometric depth of field)
Let be an acceptable blur diameter in sensor units. This is a viewing or measurement criterion, depending on sensor size, output magnification, display and task. The geometric depth of field of the fixed camera is the set of object distances with .
To describe this set, define for the fixed camera above
Proposition 1.21.5
(Depth-of-field interval)
The criterion is equivalent to . For the acceptable object distances are
For they are . At only qualifies.
Proof
(Solving the double inequality for positive distance)
Since and , we have exactly when , that is . The inequality reads . If the right side is not positive, so it holds for every ; if it is equivalent to , and one must not divide by without considering its sign. The inequality reads , equivalently . Combining these with the model restriction gives the stated sets. The lower endpoint is included when ; if , the restriction excludes itself.
For the far bound disappears: all sufficiently distant objects satisfy this chosen geometric criterion, while the near bound and remain.
Concept lensAlgebraic
Blur is linear in reciprocal distance
Using and , the blur diameter becomes
So defocus is proportional to the difference of reciprocal distances. This explains the asymmetry of the interval: equal steps in correspond to ever longer steps in as the object recedes. It also shows the far limit .
Common mistake
(The sharp range is not symmetric about the focus distance)
It is tempting to say that a camera focused at keeps points sharp up to the same distance in front of and behind . The interval above is , and is smaller than for . For mm and , the range extends about mm in front and mm behind.
Worked example 1.21.6
(The threshold changes the answer)
For the previous camera choose mm. Then ; the acceptable interval is approximately mm. A point at two meters has mm and fails this criterion. Halving the aperture diameter to mm with focus unchanged halves every geometric and doubles ; the new bounds are approximately and mm. The change in acceptable range is a consequence of the chosen threshold, not a new exact focus plane.
Worked example 1.21.7
(A tolerance above the limiting blur)
Keep , , mm and choose mm. Then . The limiting blur of an infinitely distant point is mm, which is below mm, so every sufficiently distant point passes and there is no far bound. The near bound is mm. Check: mm, as it must be at the boundary. This is the case where the criterion alone does not bound the far side.
Counterexample mode
A smaller aperture does not make every image sharper
The geometric formula decreases linearly with at fixed focus and depth, but taking toward zero does not make a real image arbitrarily sharp and bright. Diffraction grows relatively important at small apertures, and less collected light can require a longer exposure or higher gain, bringing motion blur or noise. These effects are absent from the ray model. Thus the formula supports a conditional geometric conclusion, not an unconditional optimum aperture.
From a blur disc to a sampled image
A disc boundary is not by itself a complete image formation rule. In a simple incoherent, locally shift-invariant approximation, a small image patch is the convolution of an in-focus irradiance field with a normalized point-spread function, followed by integration over each pixel and sampling. Its shape and weights depend on pupil transmission and optics. Defocus generally varies with depth, and occlusion means a single spatially invariant convolution for a whole scene is usually unjustified. A bright background point hidden by a foreground object cannot simply contribute through that foreground.
Focus errors and resampling errors are also different. A sharp scene pattern can alias when sampled too coarsely; blur can suppress some high frequencies but cannot recover lost information. Conversely, interpolating a defocused image at a finer grid does not restore unknown high-frequency content. A calibrated pinhole warp can still align approximately focused features while failing to model a depth-dependent blur. Real exposure, pupil shape, vignetting, lens distortion, aberrations, diffraction, sensor noise and focus-dependent calibration all require additional models or measurements.
Practice with complete solutions
Quick check
(Keep mm and focus instead at mm. Where is the sensor? What is for an object at mm if mm?)
Recompute for the new focus distance, then use the first form of the blur formula.
Solution · Answer
The sensor is at mm. The point at mm has image distance mm. Thus mm. Refocusing changes both the sensor position and the blur.
Quick check
(With the original , , mm camera, find the limiting blur as . Does a threshold of mm include all sufficiently distant points?)
Let in .
Solution · Answer
The limit is mm. It exceeds mm, so the threshold does not include all sufficiently distant points. Equivalently yields a finite far bound.
Quick check
(At fixed , focus and object distance, compare the geometric blur for f/2 and f/4. Does the formula alone determine the best exposure time?)
Express the aperture diameter through and ask which quantities the formula contains.
Solution · Answer
The aperture diameters are and ; the latter is half the former, so geometric is halved. The formula says nothing about scene radiance, pixel response, exposure, motion or noise, and cannot choose an exposure time.
Quick check
(For , , mm and mm, check whether points at and mm meet the criterion.)
Compare each distance with the interval, or compute directly.
Solution · Answer
Here and the interval is mm, so fails and passes. Directly, mm, while mm.