The human eye is able to see even when high brightness differences are present. This is because of the nearly logarithmic sensitivity of the human eye. Technical film footage and image sensors used in digital cameras mostly have a linear sensitivity and thus big problems on scenes with high brightness differences.

 

A technical sensor should feature a high signal-noise-ratio to be able to replace or outperform the human eye. The logarithmic measurement (decibel dB) suits the requirements to measure dynamic. Video and CCD cameras of the 90s of the last century had about 54dB, educible by 9 bits. Up to date CCD cameras from sony have about 66 dB (11 bit) and sensors from the Fujifilm company achive even a four times higher dynamic range (about 80 dB - 13 bit) by the use of a combination of high sensitive and low sensitive sensor elements.

 

                             

Comparision of normal CCD with 4. generation super CCD with extended dynamc from Fujifilm.
Image source: http://home.fujifilm.com/products/digital/ccd/pdf/web4thsuperccd.pdf

 

CMOS sensors exist since the mid 90s. Its sensor elements have a logaritmic sensitivity. The achievable dynamic of 120 to 130 dB (20 bit) outmatch the potential of the human eye whereas only 10 bit are neccessary for the data transmission. These interesting sensor elements have one problem: they have highly differences in the characteristic line of the sensitivity of each sensor element. Only corrections of each sensor element make it possible to achieve realisable images. Known correction methods for these kinds of logarithmic sensors require presumptions for the sensor elements and are extreme complex. Not so the method of the JENTECH engineering GmbH which is applied for a patent. Quickly and optimal conclusions are achieved whithout presumptions and with a self-setting, at least possible correction data amount. A software realisation of the method takes less then 200 milliseconds for the correction of a whole image with one million sensor elements. A hardware solution, which is yet to expensive, would take no actual calculation time. It would only delay the process of date reading about a few nanoseconds. With this you can achieve a frame rate up to 30 images per second with a dynamic range of 6 to 7 decades.

 

uncorrected image                                                  corrected image