sign in sign up
<<
Home , Camera lens, Image circle

Makeit

Tilt-shift

Julian Bucknall explains how special and clever post-processing create distinctive effects

hen I go on holiday, simple affair comprising two What’s covered I always take pictures standards or planes – one Distributions W of buildings in the holding the fi lm plate (and a PHOTOGRAPHIC places I visit. It doesn’t matter ground glass viewfi nder) and The random numbers your programming language run-time EFFECTS where I travel to, there’s always the other holding the . In produces come from a linear (or In modern point-and-shoot and some beautiful architecture between the two was a uniform) distribution: all values DSLR , the axis of the lens just waiting to be appreciated that didn’t let in . The are equally likely to be produced. is fi xed. If you allow the lens to and photographed. standards were mounted on Sometimes you need random be shifted or tilted you can numbers from other distributions, achieve some quite clever eff ects, The problem is, to get a one or more rails so they could such as exponential, or Poisson, like taking photos of buildings whole building in a shot, you be moved closer to or further or Gaussian. The Gaussian (or without converging verticals, or either have to move further from each other to the normal) distribution is important creating landscape photos that look away than you’d like, or point image, and to provide other in statistics. The heights of men like shots of scale models. You don’t of the same age in a population need specialised equipment though the upwards in an options like and follows a normal distribution. – some of the neater eff ects can effort to get the top in as well depth of fi eld. Figure 1 (p73) Asking a group of people to be done in your favourite photo as the bottom. When you do illustrates the basic setup. measure something with a ruler processing program. Find out how. the latter, the verticals in the Notice that the image is seen will produce values that follow a normal distribution, and so on. photo converge upwards and upside down in the viewfi nder. There is a standard algorithm the building appears narrower DSLRs reverse the image so called the Box-Muller method to at the top than at the bottom. that the view you see through generate random numbers that One solution to this issue the viewfi nder is the same follow the distribution, but taking our cue from the Gaussian blur is to use a wide-angle lens, way up as the real view. trick, a simpler method is to but this isn’t without its own generate six uniform random problems: this type of lens will Rise and fall numbers and take their average. tend to include more of the Since the bellows were The resulting averages follow a normal distribution. foreground, even as it provides fl exible and the lens and fi lm a better shot of the building. standards were separate, these The better answer is to use view cameras could provide a (known as rise and fall), or either a or a whole set of effects that are moved from side to side (the -shift lens on a DSLR. hard to produce with modern traditional shift), all while Back in the very early days point-and-shoots or DSLRs. keeping it in the same plane. of photography (and even The fi rst of these were Modern tilt-shift lenses on today with specialist cameras), shift movements. The lens DSLRs provide much of the the camera itself was a very could be raised or lowered same functionality, but they are specialised pieces of equipment and are generally very expensive compared to normal prime or zoom lenses. Spotlight on... Threshold blur The rise movement on a A Gaussian blur attemps to is in the portrait photo industry: of the edge are averaged with view camera is important for recreate the eff ect of viewing a blemishes are easily removed by similar on their particular architectural photography. To through a ground this digital process without the side), but in between the edges properly use the rise shift, the glass screen. The blur may be resulting image looking too the pixels get smoothed out since image circle of the lens must symmetric and pleasing, but it ‘Photoshopped’ and blurry. they are surrounded by similar be larger than the rectangle doesn’t preserve the edges in This works by applying the pixels. Blemishes are therefore covered by the fi lm. The image the photo; those get blurred standard box blur (or Gaussian removed without losing the circle is the area of the fi lm and smoothed out along with blur), but only using those pixels sharpness of the edges. plane formed by the cone of everything else. in the box that are similar in value Tweaks to the blur include light that comes through the Sometimes you’d to preserve (similar to within a threshold the threshold value (the bigger lens. The majority of the image those sharp edges, but blur areas value, hence the name of the the threshold the stronger the circle is masked off, leaving of the photo that look roughly blur) to the central . Edges smoothing action) and the radius just a rectangle that exposes similar. The best example of this are preserved (since both sides of the underlying Gaussian blur. the sensor of the camera (or the frame of fi lm). If the

72 321 May 2012 www.pcplus.co.uk

PCP321.theory.indd 72 3/27/12 4:40:32 PM Tilt-shift photography Make it

– also tilts (geometrically it intersects the previous two at the same position). In fact, by varying the tilt, you can bring nearer and farther parts of the view into focus at the same time (the effect on Figure 3 would be to bring all of the text into focus). Notice though that this is not the same as increasing the depth of fi eld by reducing the size of the ; this is a totally different effect, although a Tilt-shift photography photographer can balance the need for smaller to increase the depth of fi eld by using a tilt movement. Model behaviour Tilt movements are used in two contexts: where you would prefer having more of the view

▼ Figure 1: The basic schematic of a view camera in focus, and diorama effects. The diorama effect is a camera’s lens is shifted, a (just try that in the National the more obvious this sharp process that converts a different part of the light cone Portrait Gallery). Another is focus plane is. In Figure 3, photograph of an ordinary exposes the sensor or fi lm to photograph the painting I took a photo of the word ‘tilt’ view into one that looks like frame. If the image circle is from a slight angle, making in the Oxford Modern English it was taken of a scale model. small enough, the fi lm frame sure that the incident light is Dictionary at an angle with Figure 4 shows this effect on will not be fully exposed to the behind you so that refl ections the aperture wide open to a line of beach huts, and you light and you will get the effect go away from the camera, illustrate the focus plane. may also have noticed the known as , where and then fi x the perspective in There are four planes here: the diorama effect in the titles for the photo fades and darkens your favourite photo editing plane of the sensor, the plane the BBC TV series Sherlock. towards the edges and corners. program (Photoshop has of the lens, the focus plane The reason this effect works Figure 2 shows what such a feature). Alternatively, (which is parallel to the fi rst is that, through to happens with rise shift. In you can use a left/right shift two), and the plane of the page photos of miniature scenes the fi rst image (I’ve fl ipped movement to remove the in the book (at an extreme and landscapes, we know how the view so we’re not looking refl ection of the camera using angle, almost perpendicular to the depth of fi eld operates at it upside down), we see the the same technique as for the fi rst three planes). Notice for small scenes versus large normal state of affairs: the architectural photography. how the focus tapers off scenes. For miniatures, the image circle encompassing quickly the further or closer depth of fi eld is very shallow, the fi lm frame with the top Tilts and swings the text is to the camera. even at small apertures. of the building cut off. The The other main set of possible With a tilt movement, the This means that, even with foreground is also prominent. movements with a view lens is tilted up or down with small distances closer to or The second image shows the camera are called tilts and respect to the fi lm plane – further away from the camera, effect of the rise shift: the fi lm swings, known collectively as they are no longer parallel. the depth of fi eld falls off frame is in the same place tilt. In normal use, the axis The plane of the lens now dramatically. Take a look at (we’re moving the lens up and of the lens is perpendicular to intersects the plane of the fi lm Figure 3 again to see the down, not the fi lm standard), the plane of the fi lm or sensor. – usually below the camera. effect. For landscape photos, but the image circle has moved Even shift movements obey More importantly, from some even when the aperture is wide down. A different part of the this simple rule. The result of simple geometric proofs open, the depth of fi eld is view is captured by the fi lm: this is that the sharp focus of a known as the Scheimpfl ug usually quite large; more of the the one that contains the top photo is a plane parallel to the principle, the focus plane – view is seen as being in focus, and bottom of the building. fi lm. The wider your aperture, the part of the view in focus even though the focus plane is I’m sure you can imagine the still well defi ned and shallow. issues posed by vignetting if A simple way to create the image circle were smaller the diorama effect, therefore, than the one shown. is to trick the eye by blurring the region of the image closer On refl ection to and further away from Side-to-side shift movements the object of interest in the are used when photographing view. This reduction of the refl ective surfaces. One way depth of fi eld can be achieved to photograph paintings that using a tilt movement, or it are behind glass is to deck can be done through some yourself out in black and keep post-processing of the image. the camera behind a black Figure 4 was produced by curtain with only the lens post-processing – in this

poking through so that only case through some Gaussian ▼

black surfaces are refl ected Figure 2: How the rise shift movement works with a large image circle blurring in Adobe Photoshop. ▼

www.pcplus.co.uk 321 May 2012 73

PCP321.theory.indd 73 3/27/12 4:40:39 PM Make it Tilt-shift photography

calculate a Gaussian blur. The fi rst one lets you apply a box blur in two steps instead of one: you can apply the blur horizontally (it’s essentially then a one-dimensional ‘line’ blur), and then apply it vertically. The result is the same as applying the standard box blur. With a box blur that uses a simple average this trick isn’t particularly worthwhile, but it also works extremely well for a circular Gaussian blur. This means that you can reduce the number of calculations quite quickly – instead of being proportional to the square of the box side, it’s now proportional to the length of the box side. However, there is yet another trick up the

▼ Figure 4: Some beach huts showing a diorama eff ect produced by Gaussian blur algorithmist’s sleeve. Using the Central Limit Theorem, one of

▼ A blur is a mathematical of course, is the same as the Gaussian blur. Here the the major results of probability operation used in graphics averaging the nine pixels). In weighting of the pixels in the theory, it turns out that if you software to smooth out detail reality you would average each box follow a Gaussian or apply a standard box blur and reduce image noise. of the three colour channels normal curve (the familiar three or perhaps four times in A Gaussian blur gives the and recombine them to form bell-shaped curve) so that a row, the result approximates effect of viewing a photo the value for the new pixel. pixels close by have a large a Gaussian blur extremely though a ground glass screen. You could also average the weighting. This falls off rapidly well. Since a box blur is The simplest blur is a box pixels in a 5x5 box or a 7x7 box the further away you get from extremely quick to calculate blur. In this type of blur the to emphasise the blurring even the mid-point of the bell until (it’s a summation followed pixels in the original image more. Another technique you reach the long tail. by a division after all – no are transformed into the pixels would be to vary the weighting The problem with the weightings in sight), this of the resulting image by of values of the pixels in the Gaussian blur is the sheer makes a pseudo-Gaussian blur averaging out the values of box so that, for example, the number of calculations that very quick to calculate. neighbouring pixels. So if we pixels that are closer to the are required in order to take the nine-pixel box around reference pixel have a higher calculate the new value of Focus on blurring a pixel (the pixel itself plus its weighting than those further a pixel. In essence, each Back to the diorama effect. eight neighbours), a simple away (and thereby affect the neighbouring pixel value is Now we can simply select a box blur would be to take one fi nal value of the resulting multiplied by its weighting, narrow area or band of the ninth of each pixel value in pixel more than those further the total of these products is photograph to be in focus, and the box and sum them (which, away). These types of box blurs accumulated, and the fi nal everything above or below that are more properly known as total is divided by the total band to be Gaussian blurred. circular blurs, since it’s the weighting. For a reasonably If we are clever we can ‘band’ distance from the central pixel sized box, the number of the blurs themselves so that If you look at a fi lm or TV show, that’s of importance, and that multiplications grows quite the closer we are to the focused especially one shot at night with can be viewed as a series of quickly (it varies as the square band the smaller the radius of distant streetlights, you’ll see that concentric circles. of the side of the box). the blur, whereas the further out-of-focus point of light have One of the best known Luckily, there are some away, the larger the radius. specifi c shapes:. They could be circular but are often pentagons examples of circular blur is shortcuts that you can use to This will give a pretty good or hexagons. These bokeh, as diorama effect. they are called, are a result of the The diorama effect can’t be shape of the aperture, and are a applied to all photos. For a specifi c type of blurring. There’s a clever sequence in the movie start the photo must be a Scott Pilgrim vs the World where fairly wide , like the bokeh are heart-shaped. a landscape, with no large Bokeh can be simulated using objects in the foreground. The a Fourier transform coupled with a box blur. In essence, both the eye would use them to gauge image and the box weightings distances, which would negate are converted by the Fourier the blurring effect. The more transform, multiplied together, eligible photos would also be and then the transformation is reversed. This removes high taken at a slightly elevated frequencies from the image (areas angle. Nevertheless, diorama of rapid pixel value changes), or tilt effects are extremely which is a type of blurring. If you addictive to make using make the box weightings a particular ‘shape’, say a pentagon, post-processing of images, then highlights in the photo will and don’t require expensive be blurred to that shape. gadgets like view cameras

▼ Figure 3: A photo showing the narrow focus plane and shallow depth of fi eld or tilt-shift lenses. ■

74 321 May 2012 www.pcplus.co.uk

PCP321.theory.indd 74 3/27/12 4:40:41 PM