Today the affordable DSLR (digital single-lens reflex) cameras have very good parameters to perform well also in technical applications. The great advantage of such solutions is in their low cost in particular. The project will be devoted to analysis of possible utilization of DSLRs for special applications. Based on analysis and measurements of objective parameters of the DSLR’s electro-optical system a calibration and characterization methodology will be initially proposed. A database of special test images will be created using this methodology. Efficient development of image processing techniques requires a database of test images for verification of the performance, optimization and other related purposes. There is a vast of available testing material for natural or artificial multimedia content. However there is insufficient number of available sources offering well documented high dynamic range (HDR) and 3D stereoscopic visual content. In this project, based on the calibration, the DSLR device will be used to create a well structured database of images with high dynamic range i.e. HDR, and a database of 3D stereoscopic images.
Measurement of camera’s objective parametres
In order to provide complete documentation of system’s electro-optical conversion properties, spectral response and OECF measurement is essential. Also linearity of digital sensor, although intrinsic and thus expected, needed to be checked.
Sensitivity of imaging system to different spectral components of light is important property determining usability of system for various purposes. As this characteristic is usually not provided by manufacturer, it was necessary to perform measurement in our optical laboratory. There are several techniques used for evaluation of spectral response. In this project, conventional method of direct narrow-band excitation was used. This method consists of exposing imaging sensor to narrow-band filtered light and reading digital values of image. Light was filtered by calibrated computer controled monochromator. Non-constant spectral density of light source was compensated for using half-transparent mirror to split beam of filtered light and simultaneously measuring its spectrum by spectrometer. Spectrometer readings were used as a reference and image values were corrected accordingly.
Measurement was performed for different combinations of camera body, lens and UV filter to show influence of individual components to spectral response. Also two different format camera bodies were measured.
Opto-electronic conversion function (OECF)
The most important characteristic of the imaging system regarding HDR photography is its OECF. There is an ISO 14524 standard describing method for OECF evaluation using test chart to be photographed with number of different exposures. Processing of data obtained during the measurement can be done by special image quality assessment software IQjudge. This Matlab-based GUI application was developed by Josef Havlin in the frame of the project based on the core presented in his Diploma thesis and can be downloaded from this page.
OECF was also experimentally evaluated using algorithm proposed by Debevec. It was shown, that in order to get sufficiently dense OECF characteristic, one has to examine HDR set with much more and closely spaced exposures than appropriate for common HDR photographs.
Linear response of the sensor to the exposure was verified using simple technique. Image sensor was exposed to the light for increasing period of time ranging from undetectable amount to full saturation. For most precise results, measurement was repeated using narrow-band filtered light at wavelengths representing each of the three color components – red, green and blue.
For calibration purposes, several sets of artificial studio scenes were shot, but main emphasis was laid on natural scenes. During the project, around hundred HDR sets was produced in variety of conditions. Most of them were shot on tripod using Promote system for advanced exposure bracketing control without human interaction. Besides being faster, this system minimizes risk of unwanted change of camera position and direction during shoot.
Suitable exposure range and density of HDR set depends on purpose of the the data. Images intended for tone mapping and subsequent printing or internet publishing require much less exposures than for proper scientific examination, e.g. for implementation of Debevec’s OECF calculation algorithm. Part of the HDR sets produced in frame of the project offer scientific quality too.
The principle of stereoscopic images is very simple. It is necessary to make two images of the same scene, which are slightly shifted. One of them is for left and second one for right eye. If you use device allowing to display left image only to the left eye and vice-versa, human brain will automatically join these two separate images into one three dimensional perception. At first pictures were made by pair of Panasonic HD camcorders, but then also Olympus E620 and Nikon D90 DSLRs were used.
Sliding attachment on a tripod was used to provide easy shift between left and right image and to avoid any other movement. During the shooting camera should be set in a manual mode, because it is important to have the same focal length, shutter speed and aperture for left and right image.
Pair of Panasonic movie cameras allowed to create pictures of dynamic scenes, because both images are taken at the same time. On the other hand because size of the cameras prevented to position them closer than 75 mm apart. As a result the shift between left and right image is too big. That is why additional adjustment in an image processing program is necessary.
If only one separate camera is used, it is not possible to catch moving scenes, so the content suitable for imaging is limited, but also shifts smaller than 75mm can be easily adjusted. Topics of the images are focused on architecture, statues, landmarks and panoramas. Imaging with tripod is on some places in Prague forbidden, in that case pair of stereoscopic images were made from hand (namely images number 47-54), though acceptable results were achieved. Using three different cameras 70 image pairs were shot for this project.
Karel Fliegel, Petr Páta, Josef Havlín, Ondřej Čermák, Tomáš Lukeš