The smart pixel
(appeared in Dec 2018)

(link to main website)

A digital camera can work with just a single light sensor, says S.Ananthanarayanan.

Megapixels, which means millions of pixels, are the units in which the quality of cell phone cameras and professional cameras are measured. The pixel count is the number of discrete points of which an image consists. It is evident that the greater the number of pixels, the finer the grain, or the resolution of the picture. Digital cameras hence use an array of silicon chips in millions to get good images

Matthew P. Edgar, Graham M. Gibson and Miles J. Padgett, from the School of Physics and Astronomy, University of Glasgow, describe in the journal, Nature Photonics, a movement back to lesser pixels, just one pixel, in fact. This is not to say that the picture consists of lesser pixels. It is the camera, which scans the object with just one light sensor and captures the information needed for a high quality image. Using a single detector would be without several advantages of the current, multi-detector capability, but it would overcome limitations of the current technology in some applications.

The change, in a sense, is back to the way electronic images were first created. Photography started with capturing a whole image, at once, on film that consisted of specks of a light-sensitive chemical. The electronic counterpart, in the nineteenth century, was a single light-sensitive device, a photocell, which received the light, in succession, from different spots on an object to be imaged and then put them together

The image was acquired by scanning the object using an arrangement called the Nipkov disk, a circular disk with pinholes in the form of a spiral. When the object was brightly illuminated and the disk was spun, it could be arranged that light from different ‘slices’ of the object was received by a single photoreceptor on the other side of the disk. If the excitation of the receptor was made to control a light source, which was viewed through another Nipkov disk, the original image would be reproduced, and a fresh image created for every turn of the disc.

This idea, the 1884 invention of Paul Gotlieb Nipkov (1860-1940), was made use of by John Logie Baird and Charles Francis Jenkins to develop the first mechanical television. We can appreciate that the quality of the image would depend on the diameter of the disc and the number of holes. As holes in greater numbers would need to be smaller, even just a pinhole, the illumination would need to be very bright. Sure enough, the arrangement was unwieldy and the quality was poor. The arrangement made it possible to transmit images, and movement, but not to capture good images.

In the meantime, the quality of normal photography improved by leaps and bounds and film with very fine grain became available. The electronics industry worked hard to match photo film by creating panels of multiple silicon based detectors, which captured the intensity of light at a large number of spots. Miniaturization and advances in electronics led to panels that created very fine grained images, and then, methods to render the images captured. The technology has advanced and we now have low cost, hand held devices that pack millions of light gathering and emitting components. And screens of liquid crystals or light emitting devices very high resolution and accuracy.

Alongside improvements camera quality, there were advances in recording and transmission of the huge data that the millions of pixels collected. Apart from improvements in the capacity of the channels of transmission, there was data compression, or codes that reduced the quantity of data. It is well known that all digital data is expressed in binary codes, or codes that work with only the digits, ‘0’ and ‘1’. Thus, the code for the letter, ‘A’ is ‘065’, rendered in binary as, ‘01000001‘ and the word, ‘CAT’, is rendered as, ‘010000110100000101010100’. We can see that there several stretches of repeated characters. These are necessary for coding the text, but need not all be recorded or transmitted, if we have a ‘shorthand’ to express the series of ‘0’s and ‘1’s more economically. This becomes all the more true when transmitting the pixel distribution of an image. If there is dark line, for instance, rendered by a series of 10,000 pixels, we need not transmit them all, we can just say, ’10,000 dark pixels’.

For all this, conventional, multiple sensor methods have limitations in certain applications. For one, there are serious limitations of the ordinary panel of detectors in wavelengths outside the visible range. For another, very high sensitivity and speed photography are not possible with the miniaturized detectors. The development reported from Glasgow addresses these needs, by making use of a piece of hardware called a Spatial Light Modulator, essentially a high speed scanning device, along with methods to compress data

The scanning arrangement consists of an array of nearly a million closely packed mirrors, some tenth of the width of human hair, called the Digital Micromirror Device (DMD). The angle of tilt of the mirrors can be controlled and the DMD can either focus light on a particular (micro) spot on the object or direct light from a particular spot on the object to the detector. The DMD thus works to illuminate or to bring light from different points on the object to the detector, in rapid succession, to enable building the image. The single pixel detector can have very high sensitivity and speed and the limitation is only how fast the DMD can scan the object.

This limitation, the Glasgow paper explains, is overcome by scanning larger regions of the object at the same time and data compressing techniques where the image is reconstructed with collection of the least possible data. The method is akin to the game of ‘twenty question’, the paper explains. In ‘twenty question’, a player is allowed to twenty questions, of the kind that can be answered with ‘yes’ or ‘no’, to identify a person the other player has in mind. If one starts by asking, “it it a man?” for instance, the answer narrows the field to half. Further questions, of qualities that are shared by nearly half the target group, would repeatedly reduce the field to half. Doing this twenty times reduces the field a million-fold.

Using computational techniques that work like this, the limitations of the DMD are overcome and the single pixel camera has been shown to work in unusual situations. One of these is the detection of leaking methane, an invisible gas. Methane absorbs light at a particular, low, Infra Red frequency. The single pixel camera was used to display a methane pipeline, illuminated by light alternately of this specific frequency and then another one. The result was a pair of images, one showing the leak (when there is one) and the other with no leak.

Do respond to :