Bob Newman on how modern cameras shake their way to high resolution


How the eight separate images are exposed. The white square represents one pixel in the image projected on the sensor. Four exposures are made, so that an exposure is made through each colour filter in the Bayer array. Then the cycle is repeated at a half-pixel interval, effectively doubling area resolution

New features are sometimes like buses – they all seem to appear at once. Recently there has been a spate of cameras offering increased resolution using a process that involves making microscopic movements of the sensor between multiple exposures and then composing the resultant frame into one with increased resolution.

The first exponent of this technique was Hasselblad, which used sensor shift in the H5D-200c MS to produce (apparently) a 200MP output from a 50MP sensor. Four exposures were made, cycling round the Bayer sensor pattern, to take a full-colour sample at each pixel position as opposed to the single colour at each position that the raw Bayer image gives. The effect of this is a full-colour image that is especially useful in eliminating colour moiré effects, which cameras without anti-aliasing filters are prone to.

Subsequently, camera manufacturers using sensor-shift in-body image stabilisation realised that the stabilisation mechanism could be used to produce a similar effect.

First was the Olympus OM-D E-M5 Mark II, which added to the Hasselblad scheme an extra four exposures, covering the same cycle but shifted by an extra half pixel. Thus, the E-M5 Mark II, in its ‘high-resolution’ mode, is sampling the image in all three colour channels at 32 million locations, as opposed to sampling in one colour at 16 million locations in its native mode. From this it creates 40MP JPEG files, or raw files which can be converted to 64MP images.

Next was Pentax, which built into the K-3 II a pixel-shift mode rather like the original Hasselblad version. It samples full colour in each pixel location, but outputs at the sensor’s native 24MP resolution.

So, does the Olympus sensor-shift high-resolution mode produce the same result as a 32MP sensor? Obviously, the technique has its limitations. The most obvious one is that it takes longer to capture eight images than a single exposure, so the subject must not be moving. The second issue is that the pixels are too large to sample at the final pixel pitch. Rather than taking a precise sample of that pixel, they are averaging four output pixels, centred on the final output pixel position. The effect of this is a blurring of the high-resolution image. Another factor is that while sampling all colour channels at each position is advantageous, every reproduction chain in actual use ‘decimates’ the colour channels, so the final colour resolution may not be as expected.


The Olympus OM-D E-M5 Mark II includes a 40MP high-resolution composite mode

That said, for static subjects, this high-resolution mode offers real advantages, such as freedom from colour aliasing. With the fashion for cameras without anti-aliasing filters, distracting colour aliasing can wreck many images in the places where high resolution is desirable, like in the rendition of subtle textures. With the absence of these effects, the high-resolution mode is ideal for such situations.

Bob Newman is currently Professor of Computer Science at the University of Wolverhampton. He has been working with the design and development of high-technology equipment for 35 years and two of his products have won innovation awards. Bob is also a camera nut and a keen amateur photographer