As noted in a previous post, in 2015, I attempted to make a large format camera scanning back using old Canon scanners. I was inspired by this Make article. However, in the previous attempts, all I managed to do was destroy 2 otherwise perfectly good (though largely obsolete) scanners. Here are a scan:
The black band is an unresponsive element, and the light green band is an oversensitive element. These two are probably related.
A good scanner camera should be portable, and not utilize external power. This limits the choice of scanners to portable document scanners and the LIDE series of scanners from Canon.
To get started, the scanner is opened up. The two sides of the scanner are plastic strips taped onto the scanner glass. After peeling them off, the scanner glass can be slid off the casing. This gives you access to the insides, which consists essentially of a guide rod and a scanner bar assembly, which in turn includes a motor, and the sensor itself (in the black housing). The first thing to do is to disable the light source, which is used for document scanning. To do this, remove the long, clear plastic light guide rod.
Next, the LED can either be removed or covered up. Initially, I used a blob of black epoxy to cover it up.
Initially, I had wanted to make minimal modification to the scanner. Later on, I decided to just cut the LED off the PCB entirely. I had thought that using a diffuser screen onto which the lens can project will improve things, since the light will be scattered in all directions, including perpendicular to the image plane. Indeed, it does, but the improvement was nowhere near sufficient. In fact, the image above was after the addition of the diffuser screen. A look at the image circle reveals why – in the absence of a fresnel lens, the image on the diffuser was clearly circular, with significant drop-off around the edges. This means that the light was in fact scattering only slightly into the correct direction.
Recently, I’ve been thinking about this project again, after having been gifted another scanner by a friend who knows I like to tinker (Thanks Steven!). Having looked at the previous attempts, I narrowed my failure down to two things:
- Attempting to remove the sensor PCB from the plastic sensor housing, so that I can modify it to increase light intake, is extremely risky. The long, thin PCB has a tendency to bend, and the sensor elements are easily damaged. Each scanner bar has something like 8 linear sensor elements, which, when broken, renders the camera inoperable.
- Not thinking through the calibration process properly.
Making a scanner camera involves trying to project light from the camera lens directly into the sensor array, without any intervening optics. In principle this ought to work. However, the sensor array is recessed from the top of the sensor bar, and light is transmitted through a slit, via light pipes, to limit the direction from which light can enter the sensor. Under normal operation, where a light source is mounted on the sensor bar, this works well, since the direction of light between source, to the document being scanned, and back to sensor is always well-defined. On a scanner camera, however, the light from the camera enters at an angle. This angle is shallower (i.e. less ‘perpendicular’ to the scanning plane) the closer to the edges, and the shorter the focal length of the lens. At some point, the slit excludes the light from the camera. This results in a hotspot in the middle, and ugly vignetting on the left and right. This was also what necessitated the sensor bar modification.
This time round, I made sure to be extra careful when removing the sensor bar. Instead of pulling it out, I made sure to cut off all the retaining tabs, and test for free movement of the sensor elements. Then I gently pushed the bar out using a screw driver. Next, I used a Dremel and cutting wheel to cut out the plastic sensor bar.
I pretty much destroyed the sensor casing, keeping only the tabs to mount back onto the rest of the assembly. You can see that the sensor elements are now wide open – this allows light to enter from any angle.
I use the Scanner cam with my Macbook, controlled by the Vuescan software, which was purchased with this project in mind. In the settings, it is possible to perform a calibration procedure, which corrects for lighting evenness and sensor element responses. For the scanner cam, I expect that this will also be useful for setting the gain on the sensor. However, the actual process of the calibration is not completely clear. Once the calibration process is started, the software has a status bar that indicates the progress of the calibration, which happens in two steps. The black level reading is easy enough to achieve – simply put a large, black opaque sheet over the entire scanner. This is followed almost immediately by the white level reading, which I think sets the maximum level of light expected. The problem is that the white level is hard to determine. Should I be using room/environmental lighting? That will surely over-estimate the maximum amount of light the lens will let in.
If I were to use the camera lens on a camera body, exposed to a bright light, it might be better were it not a scanning back. However, since the sensor bar is parked all the way on one side of the camera, using a camera lens will surely result in vignetting in the corners, and this in turn will set the gain higher on the edges (since the corner received less light, it will have a higher gain setting). The resulting image will have bright edge bands running parallel to the direction of travel. The solution, currently, is to use a diffuser and ND filter together to bring the response to a reasonable level.
Another problem I was not expecting with the calibration was how sensitive it was to the environmental light. You can see in the image below that a ribbon cable (top right corner) is connecter to the PCB. It turns out that light reflection off the white ribbon cable was registered as more intense light by the sensor, and a consequently lower gain in those elements. This resulted in very uneven fields
The solution was to tape a black cardboard strip along the sensor, to minimize errant reflection.
The Scanner Cam is now finally ready for action. I have a 4×5 Wanderlust camera that I strapped directly onto the scanner.
First, a flat field was taken by putting a sheet of paper over the lens. This will provide information about light leaks and calibration quality.
The flat field looks pretty decent. Next up, and image was taken.
I have to say, for a first scan this is not half bad. The camera was not focused, and there was light leaks all around. But all things considered this is a decent image. I tried to correct the image using the flat field, but was not able to.
After performing the correction, the image looked worse. However, I believe I have identified the problem – neither image was controlled for in terms of exposure, contrast, etc, so the correction was not valid. After the scan is performed, I will have to save the uncompressed raw image, rather than the aesthetically better, but modified image.
The scanner cam project is officially a success. I will be mounting a new lens on a cardboard box, yielding a 32cm x 22 cm camera. I will post an update once the build is complete!