Hot pixel/column remapping in software: DNG 1.3 hack

[BenD]

Advertisement (gone after registration)

I recently noticed a relatively subtle column defect on my 1-year-old M9-P. Reviewing previous shots in more detail it seems to have been around for 3 weeks or so, affecting about 5% of shots in that time.

 

Rather than go without my only M body for a long period while Leica remap it, I looked for easy software solutions to do the same thing. In 2009, Adobe released the DNG 1.3 specification, which includes support for image correction opcodes, among others a "FixBadPixelsList" opcode which is meant to take the work of mapping pixels out of the camera firmware and into the DNG converter, which can theoretically apply more sophisticated interpolation algorithms.

 

However, I couldn't find anything which would enable me to specify a bad pixel map for my affected DNGs, so a I wrote a small Python script to do it. The source code is available here. Only a command-line interface at this stage, and some trial and error with coordinates is necessary because you have to specify the offsets in pre-demosaiced, Bayer array coordinates, but it doesn't touch image data at all and therefore preserves as much image quality as possible.

 

So far it works very well with Lightroom and ACR (sadly no support for DNG 1.3 opcodes in Aperture). The column defect is completely erased with no noise reduction and no discernible loss of detail in the area. The DNG files aren't damaged at all but I'm keeping the originals just in case.

 

If you're familiar with command-line tools and installing Python packages you might find this is a convenient way to fix affected files in bulk, rather than using Photoshop actions.

[Advertisement]

The Leica Q3 is available for pre-order at B&H Photo Video and Adorama

[tobey bilek]

How does one get the exact pixel coordinates?

[bocaburger]

The Epson R-D1s had simple and effective mapping utility in the menu, and Epson included it in a firmware update for all those who owned an original R-D1. Personally I think the omission of such a utility from Leica is ridiculous, and given that the M9 soldiers on as an in-production model in the form of the M-E, Leica should issue an update for M9-M9P-ME that includes a remapping routine. I can appreciate that remapping is a profit center for the service department, but more than the cost, the inconvenience and downtime is hugely annoying.

[BenD]

How does one get the exact pixel coordinates?

 

I used Photoshop and ACR to open an image exhibiting the issue with all noise reduction turned off. I increased saturation and lightness until the column defect was easily visible, then opened the Info palette and moved the cursor over the column of red pixels. The X-coordinate shown was the one I passed to the pixelfix utility.

[BenD]

The Epson R-D1s had simple and effective mapping utility in the menu, and Epson included it in a firmware update for all those who owned an original R-D1. Personally I think the omission of such a utility from Leica is ridiculous, and given that the M9 soldiers on as an in-production model in the form of the M-E, Leica should issue an update for M9-M9P-ME that includes a remapping routine. I can appreciate that remapping is a profit center for the service department, but more than the cost, the inconvenience and downtime is hugely annoying.

 

It does seem a strange omission, given CCDs can potentially exhibit much more severe defects than CMOS sensors.

 

I don't think it could be much of a profit centre for Leica, since many defects are time-consuming to reproduce, and a reasonable proportion of these issues would be appearing under warranty.

 

Even if a proper remapping interface couldn't fit in the camera's firmware, there must be a straightforward method available to Leica service technicians to do the remapping — if I were designing the camera, I'd do it via a simple text or binary file with a certain name saved to the SD card, similar to the firmware update process. You'd think that process could be documented and exposed to end users… or maybe someone could reverse engineer the firmware itself and publish the procedure?