- A decent film transfer begins with good exposure and white balance. The automatic systems, as used in most cameras (standard DV or HD but also machine vision), can not deal with this, because they are not made for this specific purpose.
- Een goede filmtransfer begint met een correcte belichting en witbalans. De automatische systemen die ingebouwd zijn in de meeste camera's (zowel standaard DV, HD als machine camera's) zijn niet gemaakt voor deze specifieke toepassing, en geven geen bevredigende resultaten.
-Frank Vine has analysed this problem very careful and he has developed a very special solution. It's a combination of hardware, software and a decent machine vision camera with trigger input.
-Frank Vine heeft dit probleem zeer grondig bestudeerd en ontwikkelde een zeer speciale oplossing. Het is een kombinatie van hardware, software en een goede machine vision camera, met trigger ingang.
the CFLed system by frank vine
The basic principle is to optimise the light balance and intensity entering the camera such that all three channels exactly fill the CCD pixels to the same level where there is white in the image. On an 8-bit output for example, the R, G & B levels would all be close to the 255 clipping level.
This is because the CCD dynamic range is the weakest point in the chain and we need to squeeze as much out of it as possible in order to pull out detail in the blacks without burning out the highlights. As an example, most halogen lamps put out much less blue than red. If we made a capture such that the red was at 255, the blue might be down at half that level, say 128. We would have to compensate for white balance by winding up the blue gain in the camera (or in post) to 255. This would amplify the relative noise level in the blue channel and, if this adjustment is made after the A/D, will also cause the blue channel to have a much coarser greyscale. This would be made even worse if we apply gamma to lighten the low to mid tones, which is usually required for film transfer.
With a halogen lamp, we can easily add a blue filter in front of the lamp and this does help a lot. But all films are different so we decided to make a light source which could be varied per film and scene to help optimise the capture. LEDs seemed the obvious choice, but they have their own set of problems which needed to be overcome.
The first is that they change colour frequency output with the current going through them. This makes tracking a problem when we want to change all channels together to adjust exposure.
The second is that they change output intensity with temperature and this depends also on the LED colour. Reds drop output by nearly twice the amount of Greens and Blues with increasing temperature.
CFLed is a design to overcome the above problems. It makes use of Mode1 external trigger in the camera. In this mode, not only does the camera trigger on demand of course, but the exposure length is determined by the trigger pulse length.
CFled is triggered by the projector. CFLed turns on the R, G & B LEDs and starts the camera trigger. Colour balance is achieved by adjusting the pulse time that each LED is on for. So if we want less red for example, the red pulse duration would be made shorter relative to green and blue. The camera exposure time is whatever is the longest of the three channels. So if we needed on-times of 1ms, 1.5ms and 1.2ms for R, G & B respectively to get correct exposure time and white level, then the camera shutter would be open for 1.5ms (the green exposure time). However, as the red and blue LEDs are on for shorter times, their exposures that the camera "sees" are shorter also.
This solution gets around the colour change with current in the LEDs because the current is always constant, we only change how long the LED is on for.
This "flashing" technique also helps considerably with the temperature problem because the LEDs are on for a relatively short period of time. For example, 10fps capture rate means a trigger every 100ms. Exposure times of 1 to 5ms is usual for most films with the lens set at its optimum aperture of f4. A period of 5ms in 100ms means an on-time of only 5% and hence only 5% power in the LEDs (compared with static LED lighting) and consequently only 5% heat and much less temperature rise. Nevertheless, it is still required to compensate for some temperature rise, so the LED module has a temperature sensor onboard. The system reads the temperature and automatically changes the exposure length of each channel to compensate.
The first versions of the design are stand-alone, in that all the LED control is done in hardware without any connection to the PC, other than the Firewire for frame capture of course!
The new version being worked on is a hybrid. The LED drive and pulse timing is done in hardware. The hardware has an additional USB interface to the PC. The PC reads the temperature sensor and has user control of exposures and white balance. This allows for some automation of White Balance and exposure. See here for a screen grab of a prototype in action, the idea is to balance at the start a new film/scene: auto white balance example clip
Adjustment is made by hovering over the control and using the wheel on the mouse. Hovering over the rotary knob give fine adjustment, hovering over the bar below each knob gives coarse adjustment.
There are still hardware adjustment controls as an option, these are read by the PC and change the on-screen ones instead of (or as well as) using the mouse.
Dcam2Digits is the first version of capture program and has no "knowledge" of the lighting system. The new application being worked on combines the capture capability with CFLed control.
- With special thanks to Frank for sharing this all with us and for all his work he has done so far.
More information about Franks project: The Cine2Digits Project