There are a lot of ongoing discussions on the VR Apple list about HDR capable cameras and the missing applications to exploit these files. The posts are sometimes contradictory, could you help clarify things by sharing the results of your research and consultancy work?

Well, I can try; it can be a confusing topic. I think that a lot of the confusion is due to the terminology. HDR means High Dynamic Range; this refers to the contrast ratio between the brightest point or pixel and the darkest point or pixel in a real life scene or image.

I think that there are two reasons for the confusion. One reason is that the term "dynamic range" can apply to a lot of things. The original real life scene may be high dynamic range. The camera has the ability to capture a small slice of this range. Then your printer or monitor has an output dynamic range. Image file formats also have range limitations. Images with 8 bits per color channel (often referred to as 24 bit images) such as JPEG are a Low Dynamic Range format. Sixteen bits per color channel images (a.k.a. 48 bit images) are Medium Dynamic Range images. High Dynamic Range formats are 32 bits per color channel and/or store some of the information exponentially. Thinking about the dynamic range of a scene, capture device, image, file format, and output can get confusing.

Another reason for the confusion is because most photographers have only one goal they would like to achieve - capture an HDR and make it viewable on a low dynamic range display, such as a print or monitor. They may start with a scene that has High Dynamic Range and then take exposure-bracketed images that are all low dynamic range. If they then use an application like Photomatrix, or composite it in Photoshop, they never truly have a HDR image. A point that I think a lot of people miss is that once you composite your images in Photoshop, Photomatrix or even tone map a true HDR image file, you no longer have a high dynamic range image; the range has been compressed into a low dynamic range so that it can fit into an image format such as a JPEG and displayed on your monitor.

This type of automatic process that would compress the range of an image down to that of our monitors and printers (and ideally to our color sync and ICC color profiles) is a holy grail in photography. The range of film has limited photographers since the beginning of the technology and it looks as if a good automatic solution is right around the corner. However, I would say that none of the solutions I have seen so far are robust enough or of high enough quality to depend on for production work. For now, I would say the best results are still achieved by careful Photoshop work; I hope that is about to change.

You were an HDR imaging consultant for Spheron. What are the most interesting and performance application fields for their camera?

I had an opportunity to use their camera for a few weeks when I was living in France and helping them with some of the potential applications.

Spheron's SpheroCam HDR has some really clever engineering and a lot of little features that make you feel like you're using a well thought out product. I was quite impressed at the range that the camera could capture in a single scan, it's greater than what you can capture with an SLR by bracketing. In most situations, you can easily capture the whole range of the scene, with the exception of the sun, as it is much too bright. At f-8 you would have to photograph the sun at 1/100,000th of a second to reduce it to a circle and get color information and this is beyond the capabilities of normal cameras.

So far Spheron has only addressed the very high end with this camera. Most people who are using it are using the HDR images for lighting 3D. I think that it's likely that this will be the primary use for this camera in the near term, even though there are several other camera effects you could achieve, such as realistic motion blur, blooming, exposure adjustments etc.

One of the largest untapped potentials of HDR is using it as textures for 3D models. For example, you can have dynamic lighting that interacts with the environment (HDR panoramas can only define the lighting for a given point in a scene). Imagine you are making a movie and have a synthetic character, a 3D generated alien. If you want this alien to walk around a room and use HDR to light it you will have to plan in advance where you want the character to walk and then take an HDR every two inches or so along the path he will walk. So as he approaches a red wall in the scene he will get the right amount of red light bouncing off the wall and illuminating him. This would be a lot of work and require you have a robotic rig taking all the pictures. If the director then changed his mind and wanted the alien to walk in front of the couch instead of behind it, you would have a big mess on your hands. If you instead took a two or three HDR in the room and acquired the room's geometry via my PIMP techniques or a laser scanner, and then used the HDR as textures, you could determine the lighting for any given point in the room and change the path of the character at will in postproduction. This would save time, cut down on equipment rentals in production and provide flexibility in postproduction.

Are you familiar with the new Panoscan MK2 system? It seems that the basic difference with the Spheron is the missing HDR capabilities. Can you explain what this means in terms of application fields?

You can still make high resolution HDRI with the MK2, it just means that you have to make multiple scans. Ted Chavalas of Panoscan and I made a few at the "IQTVRA Summit in the Rockies?.

The MK2 already has a fair amount of dynamic range; its CCD records at 14 bits per color channel and stores that info in a 16-bit image format. Because the pixels in the MK2 are larger they gather more light, which means the camera can scan faster. The 14 bits of range means that you will have to do fewer scans to capture a given range than if you had an 8-bit camera. I haven't tested how much range you can capture with this camera by bracketing, as the dynamic range of the location where we tested it was not that great.

Spheron's camera can scan a very wide dynamic range but it scans slowly at full resolution. So you are faced with either multiple faster scans or one slow scan. Spheron also has the advantage no postproduction, you can have an HDR on the spot; however, making an HDR from a scanning camera takes 5 minutes or less and if you needed to you could batch the process in postproduction with the MK2. In either case, it is only really practical with still life as moving objects will throw a monkey wrench into the works. The MK2 would have a more difficult time if the sun moved noticeably between the first and last scan, as the shadows would not match, producing data errors with a noticeable artifact. Since the SpheroCam HDR is capturing all the range in one scan, even if the edge of a shadow does not meet the point it started this would be a negligible artifact as you could scan more than 360-degrees and blend the difference, as I have done with my HDRI stitching technique.

Unfortunately, right now there is no good solution for environments where things are moving in the scene or the light is changing rapidly. I am hoping that this will change as more camera hardware manufacturers consider this problem. I find Fuji's "wide dynamic range" CCD an encouraging development. I just hope that as manufacturers develop these capabilities they offer the ability to hand the photographer all the HDR data via a raw format rather than doing all the range compression in the camera.

For more information on and examples of HDRI, visit the HDRI section of Downing's website.

More of Greg Downing's work:
Downing’s Garfield Trailer

Petra Exhibit at the American Museum of Natural History in New York, featuring a Surround movie generated by Downing’s panoramas

Email: greg@gregdowning.com