Since a few years ago, digital cameras and mobile phones have been fighting a marketing war with their competitors to see who can provide the greatest number of megapixels to their devices. In this way, one can find mobiles with 15 megapixels, a resolution that is similar to that which some professional digital cameras had. This is solely marketing strategy, in fact, the quality of images is limited either by the aberrations of optics (established by its quality) or by the diffraction limit.
The optical resolution of a device, in case of diffraction, depends on its focal distance and the optical aperture, not being very strict we can compare it to the diameter of a sensor.
Thi minimum size of each pixel in case of a mobile phone with a micro-camera with a 6mm diameter and a normal focal range of 18 mm, for a visible wavelength of 600nm, with perfect optics,would be 2.196 micrometers (0,002196mm). As a whole, the sensors of a mobile device are 1/6’’, which is 2.4×1.8 mm (Width x Height). In this way, assuming that each pixel has a limit size of 2,196 micrometers, a mobile’s image sensor would be optically limited to 1092 X 820 pixels (< 1 Mpx) due to the diffraction phenomena. Thus, we see that the real quality of mobile cameras is much smaller than the resolution they offer. If present-day devices are optically limited, then could we obtain images beyond the limits of optical resolution? The answer is yes. Although the resolution of a sole image is limited, there are ways to improve this resolution by extracting embedded information in the image or in its sequences. These techniques are known as superresolution. These techniques are based on either the use of the information contained in several sequential images in which there is small displacement between them or in learnt statistical models. At present, we are going to look at the information contained in several images. If we are able to locate with sub-pixel accuracy the relation between the pixels in different images, then we can generate a greater resolution mesh by correctly interpoling the information in the different images and joining them into a high resolution image. This is achieved by means of a mathematical process of optimisation which generates a real high resolution image that has a greater probability of creating images than those low resolution images available.
These techniques can be used for the improvement of microscopy images where the increase in optical quality entails a very high cost, such as histological images that are shown below:
These techniques can be used on a low resolution video sequence to improve its resolution. This means one could automatically improve the quality of existing videos taken with low quality sensors as can be seen in this endoscopy application.
In sum, superresolution techniques help us overcome the optical capture limitations of low quality devices and can be employed in numerous applications which range from medical images to improving our old VHS videos.