Our HDR display is based on the SBT1.3 model made by Sunnybrook / Brightside Technologies. It uses the same Sharp 15" colour LCD model LQ150X1DG0, but we replaced the original DLP projector with Acer P5290. The main advantage of the new projector is a much higher peak brightness, 4000 ANSI lumens, as compared to 1100 ANSI lumens produced by EzPro735 used in the original design. This, in turn, allowed us to retain the colour wheel of the DLP, which is used to change the colour of the light it projects, but lowers its peak brightness. Apart from a higher brightness, the new projector has also a much higher contrast (3700:1, as opposed to 500:1 offered by the originally mounted Optoma DLP) and a greater focal range adjustment, allowing to focus the image on the back of the LCD panel and thus reduce the amount of blur.

Software

Calibration

The calibration of the display is composed of two major blocks, geometric calibration and colorimetric calibration. Both stages are semi-automatic and require little input from the user. In the geometric calibration part, a Canon Rebel XS digital SLR camera is used to calculate the transformation that aligns the backlight with the LCD screen, while also measuring the PSF of the DLP as well as the DLP non-uniformity image. Such image is then used to compensate for the projector luminance non-uniformity during rendering. The effect of vignetting on the photographs was measured beforehand to remove its influence on the result.
Initially, image alignment between the LCD and DLP projector was achieved by finding a projective transformation of DLP image corners. This proved to be insufficient, as the DLP projector introduced a non-linear deformation that could not be compensated by the projective transformation. This is why in the final algorithm the image is rendered on a freely deformable grid mesh, with the position of each vertex transformed using the projective transformation and then translated individually to overlap with the LCD. The translations of the vertices are calculated during the calibration procedure.
The colourimetric calibration was done using a spectrometer (JETI Specbos 1211). Since the DLP produces greyscale image, its calibration consists in measuring a 1D look-up table describing the function mapping pixel values to luminance. For the LCD, a more complex model needs to be fitted, as it is responsible for adding colour to the resulting image. The gain-offset-gamma (GOG) model was used to characterise the LCD panel.

Rendering

Measurements

The measured gamut can be on the CIE xy chromaticity diagram below. It should be noted that when the achromatic (ie. providing greyscale image only, with colour added only by the LCD panel) backlight is used; the gamut of the display is much smaller than the sRGB gamut, which is supposed to represent the colour gamut of a typical display available on the market. This is caused by the fact that our backlight was not selected specifically for the LCD panel we use. We decided to determine if it would be beneficial to use a chromatic backlight instead of achromatic. We measured the gamut, but for each primary colour on the LCD, the DLP was set to the corresponding colour. The result, which is also presented below, shows that the colour gamut can be extended to be similar to the sRGB gamut if the chromatic backlight is used.

CIE xy chromatic diagram with the color gamut of our HDR display using achromatic and chromatic backlight. The sRGB gamut is also shown for the reference.

Next, we measure the contrast of our display. There are two popular methods of measuring contrast: the full on/full off measurement and the ANSI measurement. In the first method, the luminance is measured when the whole screen is set to the brightest white, then to the darkest black and the results are divided by each other. The second approach takes into account that when actual images are displayed, some light from the bright areas may increase the brightness of dark areas. Thus, to measure the ANSI contrast, a rectangular checkerboard with 4x4 tiles is displayed on the screen and the luminance is measured in the middle of any white and in the middle of any black tile. This results in a measurement that is much more applicable to real life situations. The ANSI contrast tends to be lower than the full on/off contrast.

We took an approach that provides more information about the true contrast than the ANSI measurement. ANSI contrast was created to measure the contrast of displays with uniform backlight, whose properties do not change much depending on the size of the pattern displayed. Because our backlight is provided by a projector and is passed through a diffuser before falling on the back of the screen, the problem of the light bleeding from light to dark areas is much more pronounced and varied depending on the size of pattern displayed. For this reason, our measurements included a wider range of checkerboard tile sizes, from the size that covers the full screen with a single tile to the smallest area our spectometer could cover, which amounts to 40 pixels. The results of our measurements can be seen below. As expected, the luminance at white squares is almost independent of the tile size in pixels. The luminance of black, however, changes substantially depending on the pattern displayed. This limits the available dynamic range from 5.4 orders of magnitude (17.9 f-stops, 251,190:1) in the full on/full off mode to 3.7 orders of magnitude (12.3 f-stops, 5,012:1) when a very fine pattern is being shown. We also compare these results with the local contrast offered by the LCD panel used in our HDR display. This is achieved by conducting identical measurements when the DLP provides a constant, uniform backlight.

The measured luminance of white and black check-board tiles at varying size of the check board.

The contrast of the HDR display compared with the contrast of the LCD front panel.

Photographs

Publications