Image calibration using AI

In this article we discuss image calibration, lens distortion and their use in digitising.

This stage of the pattern digitisation process is crucial not only for obtaining accurate results but also for speed and convenience.

What is image calibration and why do we need it?

Although we rarely notice when we take pictures in daily life, most cameras and mobile phones take distorted pictures, with various types and degrees of warp, which is determined by various attributes of their lenses.

This distortion becomes more significant, therefore easier to notice with naked eye when the device used to take pictures has a wide-angle lens, such as adventure cameras and doorbells. Fisheye effect is one of the most well known type of distortions.

In fact, human eyes (as well as eyes of most other complex organisms) have similar structure to camera lenses. The convex shape of our eyes provides us with a wide-angle view, which means the raw image formed on our retina also suffers from the lens effects. However, this is then corrected by the visual cortex of our brain, which interprets the neural signals into vision.

Biological lens

Lens optics

On the other hand if the image is being acquired via a scanner, there’s no such effects due to the difference in the method scanners work.

The visual distortion mentioned above does not pose any problem if the pictures taken are for leisure purposes. However if the pictures are being used for extracting important data such as measurements (which is the case in vectorisation or digitisation) that require precision then the distortions such as lens and perspective effects must be reversed in order to obtain a flat image. This process is also known as “calibration”.

The main challenge in image calibration

If we consider the number of manufacturers of cameras, mobile phones and other devices that can be used to capture images, and the number of different products and their versions, then we’re speaking about thousands of different ways these images can be distorted due to the diversity of lenses these devices might have.

Some might warp the image inwards and some outwards with varying degrees, which are called “Pincushion effect” and “Barrel effect” respectively. Some lenses can even produce a mixture of these, which is called “Moustache effect”.

Although focal length plays a role in the degree of resulting distortion, two different cameras with the same focal length are likely to produce different results, which makes it impossible to use one set of parameters for all devices.

Pincushion effect

Barrel effect

What are the traditional methods to remove visual distortions?

Today there are various methods and tools to remove lens distortion from images, such as Photoshop and Lightroom. This typically involves using a printed checkered pattern. However as these are not intended for use cases that require precision, extra caution should be taken. In addition to the lens distortions, perspective distortion should also be considered, which takes additional time and effort to correct.

Some digitising systems use a grid printed on a sheet of paper for manually calibrating the pictures, however this method has a couple of important drawbacks. First, locating the cross points of the grid manually takes a significant amount of time which is proportional to the size of the surface being calibrated. The second disadvantage is the fact that the camera and the surface must remain in fixed positions in relation to each other, otherwise the calibration would be invalid. This means that the camera can not be held by hand, which restricts flexibility severely.

How iDigit does it?

We believe that a digitising system must be user friendly and provide superior precision without

having to compromise on one for another. This requires all the nitty-gritty calibration related steps

be carried out automatically in the background.

Thanks to our unique iDigit board and the use of advanced machine learning techniques, iDigit can

integrate any camera or mobile phone into its camera library effortlessly. Once this step done, you

can use your camera or mobile phone in any way that suits you, including hand-held, tripod mounted or wall-fixed.

Once a picture is shot and loaded into iDigit, the lens distortions as well as the perspective effect

will be automatically corrected in the background in a couple of seconds, followed by the

automatic detection of patterns and their components, also in a matter of seconds.

As iDigit determines the device that was used to take the picture from the picture automatically,

you can use iDigit with several different cameras and mobile phones simultaneously, without

having to change any settings.

Here’s a visualisation (slowed down by 100 x) of the training process of the Artificial Neural Network

used by iDigit to perform automatic calibration.