Every essay — page 9
What survives Constructing a view Where to stand Surfaces that are not flat The other systems Light and mirrors Mirrors that are not cameras Measuring from one picture Through water and glass The real instrument The second eye What a pair is for Many pictures at once The eye that moves Systems that kept the measure What each system gave up What a machine computes The rectangle behind the lens The second projection Drawn confidently
The real instrument
A lens is a departure from the pinhole, and the departure is the largest systematic error in every measurement made here. It bends straight lines, destroys the invariant, and can be recovered from nothing but the knowledge that some edges were straight — which is the round trip again, on a harder problem.
A tilted sensor is not a distortion
Tilt a sensor 3° out of square with its lens and every point of the picture moves — up to 7.5 px on the frame drawn here — yet every straight line stays straight to 10⁻¹³ px and the cross-ratio survives to 10⁻¹⁶. The picture is an ordinary pinhole picture whose principal point has moved 22.30 px. A calibration that frees its principal point absorbs it exactly; one that holds the principal point and reaches for tangential distortion terms leaves 1.87 px, and used as a correction it bends straight rows by 4 px.
The lines that calibrate a lens
One straight edge through the centre of a picture says nothing about a lens's distortion, and one 180 px from the centre determines k₁ to 9.0 × 10⁻⁴ — the precision rises in proportion to the offset. But distance from the centre is not enough. Crowd three edges on one side and, the moment the distortion centre is also unknown, the coefficient is ten times worse, because a bend on one side looks like a moved centre; put one edge across the centre and it barely changes.
The entrance pupil walks with the angle
The place a picture is a projection from is not a point in a wide-angle design. Chief rays traced through a strongly curved front element cross the axis 15.07 mm behind its front vertex when they are nearly on the axis, and 4.23 mm nearer the front at 80° of field. So no pivot makes a wide panorama seam clean: at one metre, pivoting at the paraxial pupil leaves 4.39 arcminutes of misregistration along a seam, and the best pivot still leaves 1.41.
A model that inverts has a horizon instead of a fold
The polynomial every calibration fits turns around at a finite radius and stops being a map from direction to picture. The division model, chosen because it inverts in closed form, never turns around — it rises for ever toward a horizon at one over the root of its own coefficient, so the whole hemisphere of directions lands inside a finite disc. Fitted to the four fisheye laws it follows every one of them more closely than the polynomial at every field from forty degrees to eighty — a hundred times more closely for the equidistant law at forty, and the stereographic law exactly.
The response is at the ends and the information is not
A radial map bows a straight edge by an amount that grows as the square of the distance along it, so 93 per cent of an edge's response to the coefficient lies in its outer quarters. Spending the marks there is 16 per cent worse than spreading them evenly, because two clusters say nothing a shifted, tilted line could not say. What identifies the coefficient is a curvature, which needs three places — both ends and the middle, which beats an even spread by 11 per cent.
Focusing moves the pivot past its best place
Focusing a fifty-millimetre lens to one metre carries its entrance pupil 2.63 millimetres forward of the camera body, and to half a metre 5.56 — which is more than the whole 5.53 that the pupil walks with field angle, so past a subject at 502 millimetres the focus decides where the pupil is. A panorama head aligned at infinity and used at a metre leaves 7.46 arcminutes along its seam; aligned at four metres it leaves 2.21, better than pivoting at the pupil at all.
No design separates the two coefficients
Separating a squared term from a fourth-power one was supposed to need marks at radii far apart, which is a statement about where edges are placed. It is not: one straight edge already runs from 20 px to 326. Spreading ninety-six marks over three edges or sixteen changes the answer by 23 per cent, spreading the offsets makes it 19 per cent worse, and the correlation stays at −0.98 whatever is done. What a plumb-line calibration determines is one number, to 1.52 thousandths, and which number depends on the model.
A known target sharpens the fit and does not separate it
Printed circles of stated size were supposed to break the −0.98 correlation between a lens's two radial coefficients, because a circle puts every mark at one radius and no straight edge can. They do not: every design of circles leaves the pair 0.979 to 0.9997 correlated, and for circles of known size the figure is exactly the cosine between r³ and r⁵. What a known target buys is precision — 3.8 times the straight edges' at the same budget — and only if its size in the picture is known to about a thousandth.
A stereographic fisheye is a division model
The division model divides the picture radius by one plus a coefficient times its square. The fits that compared it with the polynomial were not of that model: they multiplied instead, and the model they measured has neither a fold nor a horizon. Fitted as it is written, the division model follows every fisheye law more closely than the polynomial at every field from forty degrees, and the stereographic law it follows exactly — the law is the model, with a coefficient of minus a quarter. Its horizon then turns out to sit beyond the lens's own ninety degrees, and pinning it there is a trade rather than a free constraint.
A tilted target pays for its tilt in perspective
A calibration target of printed circles held square to the lens fixes a lens's first radial coefficient to 4.77 thousandths from ninety-six marks. Tilt it sixty degrees, fit the tilt along with everything else, and the same ninety-six marks fix it to 1.70 — nearly three times better. Squash the circles by the same angle without perspective and nothing is gained. What pays is the near half of each circle being drawn larger than the far half, which spreads a circle's marks across a band of distances from the lens's centre.
An image circle pins a fisheye's horizon, if its field is known
A fisheye picture stops at a circle, and for a lens covering exactly 180° that circle is the ninety-degree radius a division model calls its horizon. Handed to a calibration as one more observation, read to two pixels, it takes the error at the edge of an equidistant lens from 9.4 pixels to 2.0 — at a cost of 0.3 pixels inside the field the targets covered. Read as ninety degrees on a lens that really covers 190°, it makes the edge five times worse than no circle at all.
A refocused panorama is six lenses
Refocus between the frames of a room panorama and each frame is its own camera: its pupil carried forward by its own extension, its picture made at its own principal distance. The best fixed pivot then leaves 12.9 arcminutes along the worst seam, and a head that slides the camera back as the lens extends leaves 2.82 — the lens's walk alone. The pivot's share is first order and worth a tracking head indoors. But a stitcher that reads all six frames with one focal length misregisters by up to 270 arcminutes, which is the larger mistake by ninety times.
One focus stopped down is as sharp as six in a deep room
Refocusing a room panorama frame by frame costs a seam of 12.9 arcminutes on a fixed head. One focus stopped down costs none of that, and at f/40 holds half a metre to four metres to 5.27 arcminutes — sharper than the fixed head's seam, blunter than a tracking head's 2.82. But the comparison flatters refocusing: the frame on the table also holds the wall behind it, so it cannot be held sharper than 5.27 however it is focused. Counted properly, one focus on a plain head comes within 2.4 per cent of the geared one, and pays for it in light.
A level picture shows its rise on its horizon, and hides its slide
A shift lens moves the principal point off the middle of the frame, and the textbook focal length from two vanishing points pays for assuming it did not. On a level camera the payment is optional. The principal point lies on the horizon, so a rise is read straight off the picture and costs nothing; a slide along the horizon is read by nothing at all, because two vanishing points fix only a semicircle of possible cameras. What a sideways crop costs then depends on how the building is turned: a tenth of the frame trimmed off the side is 23 per cent of the focal length on a facade seen eleven degrees off square and under one per cent at forty-five.
A second rectangle fixes the column a level picture hides
Two vanishing points of a building put a level camera on a semicircle and leave it free to slide along it; a second rectangular object turned against the first puts it on a second semicircle, and the two cross once, at the camera. On a facade seen twelve degrees off square and slid a hundred pixels, a second object twenty degrees further already reads the focal length to 1.3 per cent where the building alone, even with its principal point known, manages 2.9 — and where the frame's middle column is wrong by a third. The reading does not care where the column was slid, and it is best when the second object stands at forty-five degrees to the camera, whose semicircle is flat over the true column.
Light and mirrors
A shadow is a projection from the lamp; a reflection in a flat mirror is the view from a camera on the far side of it. Neither needs a new method, and both are computed with the projections already here — which is exactly why the mirrors that bend get a field of their own.
A shadow is a second projection
The construction that puts a shadow on the ground is the construction that puts the scene on the picture plane, with the lamp where the eye was. Shadow drawing is taught as a separate set of recipes and it is one operation with the centre moved, which is why the same code draws both.
Where shadows vanish
The shadows of parallel posts under the sun meet at a point, and that point must lie on the horizon. Under a lamp they meet at the lamp's foot instead. Both are checkable in any photograph, and a picture whose shadows fail the check was not lit by anything.
A mirror is a second camera
Reflect the scene and photograph it, or reflect the camera and photograph the scene. The two routes disagree by 315 px and agree to the last bit once one axis of the image is reversed — which is the whole of why a mirror is said to swap left and right, written down.
The penumbra is the lamp's image
The soft edge of a shadow is a picture of the light, projected through the occluder's edge as through a pinhole. That gives its width without any integration — and it is why the dapples under a tree go crescent-shaped during an eclipse.
The shadow of a ball is a conic
A cone cut by a plane is a conic section — which is the definition of the phrase, not an analogy — and the shadow of a ball is exactly that cut. So the shadow closes or runs to the horizon depending on where the lamp is, and the condition is not what anyone guesses: it is that the lamp is above the top of the ball. The ball's distance appears nowhere in it.
The lamp, out of the picture
Two posts and their shadows put the light's position in a photograph, exactly, with nothing given but the drawn lines and the camera's own horizon. Two posts. One gives a residual of zero and an answer that is a whole one-parameter family — the sharpest counter-example there is to the idea that a small residual means a right answer, met again in a new field.
A shadow can be un-cast
A shadow looks like a lossy record — a flattened smear with the shape half thrown away. It is nothing of the kind. The map from the occluder's plane to the floor is a plane projectivity, so it has an inverse, and the outline comes back out of its own shadow exactly. What breaks it is not the light and not the shape: it is the floor not being flat.
A wall does not get darker as it goes away
The inverse square law is about a point source. A surface is not a point source, and the picture of a wall is exactly as bright at twenty metres as at two — the patch one pixel covers grows as the square of the distance and the light per unit area falls as the square of the distance, and a picture records the product. Which is why aerial perspective has to be the air.
A lamp lights less than half a ball
Everyone knows a sphere is half lit. It is half lit by a source at infinity and by nothing else: a lamp two radii away lights a quarter of it, and the boundary it draws is a circle offset toward the light rather than a great circle. The offset is R²/D, which means a photograph of a lit ball carries the distance to whatever lit it.