After reviewing the official Sigma DP1 examples, my thoughts turned to the best method for upsizing them. While it is clear that these images contain a ton of detail, the choice of upscaling approach is important for getting the best possible extra large prints out of relatively small (4.65MP) files. As luck would have it, Ctein of The Online Photographer today posted Part 1 of a three-part comparison between Genuine Fractals, Blow Up, and Photoshop Bicubic. I'm looking forward to Parts 2 & 3, which are said to be coming over the next week.

In the meanwhile, I've been having fun experimenting with resizing approaches. In particular, I've been comparing Genuine Fractals and Photoshop Bicubic Smoother for their ability to upsize the DP1 samples to 14MP. The major problem I've encountered with upsizing the SD14 and DP1 files is that of aliasing artifacts (jaggies). Genuine Fractals seems to do an excellent job of preserving detail while minimizing these artifacts. I'm finding that the default setting on Genuine Fractals used to upsize followed by modest Smart Sharpening in PS (Amount 100, Radius 0.3) gives good results. In the comparison crops that follow, I compared this approach to Photoshop Bicubic Smoother, which required more aggressive subsequent sharpening (Smart Sharpen, Amount 200, Radius 1) for a similar degree of apparent detail. Disclaimers up front: 1) I don't yet have permission from Sigma to post upsized images, but I believe that using these tiny crops as examples is permissibile from a copyright standpoint. The original photographers for the images from which these crops were derived agreed with this; 2) JPEGs aren't the best source for upscaling. JPEG artifacts become more prominent; 3) Obviously my methods for upscaling may be suboptimal and therefore not representative of the DP1 X3 potential; 4) Rick Decker pointed out that the crop I chose from his image may not do justice to the sharpness possible with the DP1 given that the light wasn't that bright at the time of the capture.

Here are some examples (Click image for intended viewing size):


Another example (Click image for intended viewing size):

The differences are definitely subtle, and it would probably take quite a large print to show them. However, if you pick through looking for aliasing effects, they're present in the images resized using PS Bicubic to a greater extent than in those resized with GF5. The jaggies on the slope near the observatory are the most prominent example, but there are others if you look closely at 100%.

In a response to Ctein's article, Hugh Crawford posted the following insights regarding Genuine Fractals: "The behavior of the Genuine Fractals makes sense if you keep in mind that it was originally designed as a file compression program that converted the raster information to vector information and compressed it using wavelets. It tends to want to define things as areas with edges. On the one hand it does a good job of getting rid of aliasing effects on edges and sharpening them without the usual sharpening artifacts, but at the expense of wanting to make things look like areas with edges if the blur is near the sampling frequency." These comments agree well with my preliminary results shown above; and in examples not shown, I've seen a bit of the downside to which he referred.


I have written to Sigma for permission to feature upsampled images on this site. Depending on their reply, I will be able to more fully demonstrate the potential of this approach.

If you use the link below to join me in pre-ordering the DP1, Amazon will give me a small credit towards the purchase of my DP1. Thanks!

The digital camera industry is heavily invested in the concept of megapixels, where one megapixel = one million pixels. One can define the number of megapixels based on the number of pixels in the camera sensor, or one can define the number based on the number of pixels in the resulting image. With nearly all cameras, there is no substantial difference between the two. For example, Canon's flagship compact digital camera, the Powershot G9, has a sensor made up of 12 million pixels in a single layer and outputs a final image that is 4000 pixels by 3000 pixels (12MP). The Foveon X3 sensor in the Sigma DP1 and SD14 is an exception. This sensor uses three layers of pixels with 4.6 million pixels in each layer. While that means that there are roughly 14 million pixels altogether in the sensor, the output of this sensor is an image measuring 2640 pixels by 1760 pixels (4.6MP). Sigma and Foveon detractors have pointed out the obvious discrepancy here. Clearly a 4.6MP output image cannot contain 14MP of detail. For essentially marketing reasons, Sigma is referring to the DP1 as a 14MP camera. After all, who would pay $1000 ($800 street) for a 4.6MP pocket camera?

As it turns out, Sigma is forcing us to re-evaluate megapixel ratings just as AMD forced us to re-evaluate gigahertz ratings in the computer processor industry. Not all megapixels are created equal from a detail standpoint, and that's without even considering other aspects of image quality including color fidelity, noise, and dynamic range. Anyone who's ever put a mediocre lens on a good DSLR can tell you that to do so is to literally throw away megapixels. In fact, it takes a very good lens to fully utilize the resolving ability of a sensor across the entire frame. Assuming one has such a good lens on hand, the next issue is whether the sensor itself can deliver in terms of "per pixel sharpness." For technical reasons that are well explained elsewhere, the sensors used in most cameras simply don't deliver as much detail as their megapixel ratings imply. Without delving into the reasons for this, it is easy to recognize the issue. View any image from a high megapixel camera on screen at 100%, and one can see that the level of detail simply isn't as high as it is in a resized image. At their "per pixel level," these images are just a little bit blurry. Typically speaking, one has to resize the image to 75% or less and then sharpen before the apparent detail is truly crisp. On average, this resizing step corresponds to an approximately 50% loss in megapixels. Depending on the particular camera and lens combination, the resizing may have to be even more significant. For example, a DSLR with a strong AA filter and a mediocre lens may output only a fourth of the advertised MP count in terms of true detail.

After examining hundreds of Sigma SD14 images and all available DP1 samples available online, it is clear to me that the me that this Foveon sensor provides better "per pixel sharpness" than any camera I have ever owned. The quality of the detail looks as good at 100% on screen as it does resized, and that is refreshing. Rather than advertising the DP1 as a 4.6MP camera with superior per pixel sharpness, Sigma has decided to classify it based on the number of pixels in the sensor. While the latter is indeed somewhat misleading, the former would have been disastrous from a marketing standpoint. As far as I am concerned, Sigma has been as forthright as have been other camera manufacturers in declaring megapixel counts. To quote Wikipedia's entry on the Pentium 4, "a review showed that in games, it took a 5.2 GHz Prescott core to soundly beat the performance of a 64-bit Athlon FX-55 that clocked at 2.6 GHz." Thanks to AMD, most of us now know that the gigahertz specification of a computer processor gives only limited information about overall performance. Thanks to Foveon and Sigma, it is more clear than ever that the megapixel rating is a highly limited assessment of the level of detail a camera can output.

Given the available samples from the SD14, I'm satisfied that the DP1 sensor (same as the SD14) can deliver higher overall image quality than that of any other pocket camera available to date. Based on the DPReview DP1 samples, I believe that the lens will be up to the task. Therefore, despite the fact that 28mm is not the single best focal length for my overall intended use - and with the understanding that the f/4 limitation is what it is - I pre-ordered the DP1 on Amazon earlier today. Amazon initially had the DP1 available for pre-order at a price of $750. It's currently at $800. Amazon's policy is to give one the lowest price at which it is listed between the time of pre-order placement and one month after shipping. No charges are made until the camera ships. After placing my order, I was given an estimated ship date of March 31 despite availability being listed as March 25. I'd be curious to know whether anyone else who pre-ordered earlier from Amazon was given an earlier ship date, as this would suggest that they are getting a significant number of orders.

If you decide to join me in pre-ordering the DP1 and would like to tip me in the process, you can do so by using the link below to get to Amazon.

Out of interest in the upcoming DP1, I've recently been reading everything I can find about the Sigma SD14, with which the DP1 shares a common sensor. That led me to Carl Rytterfalk's excellent site. Going through Carl's archives, I found a link to "The Great 14MP Shootout - Sigma SD14 vs Kodak 14nx" by Whisperingcat Photography. David Millier and Erik Muehlberger did an outstanding job of comparing the output of the Sigma SD14 and the Kodak 14nx. Their article is of general interest to anyone who pixel peeps 100% crops to gain information about sensors and of particular interest to those looking to learn more about the Foveon sensor in the DP1 and SD14. That comparison, again, is here.

Carl Rytterfalk is posting some very interesting work on his site. He is using the Sigma SD14, which has the same Foveon sensor as the Sigma DP1. While the samples available from pre-production DP1 units (available at here at DPReview and here at LetsGoDigital) have shown impressive performance at low ISOs, the high ISO files have been criticized for "blotchy" color noise. In my experience, similar color noise from Bayer sensor-derived high ISO files typically translates into "blotchy" regions in black and white conversions. Carl's ISO 3200-12800 tests using the SD14 suggest that the same may not be the case for the latest Foveon offering. His additional high ISO tests were performed at 16mm (28mm FOV in 35mm photography terms) to simulate the expected output from the DP1. Carl recently also posted a black and white portrait from the SD14 resized from 4.6MP to 14MP. Anyone interested in the SD14 or DP1 will want to check out his excellent work. To see nearly 2,000 more examples of photos taken with the SD14, check out the SD14 group on Flickr.

The DP1 is now available for pre-order from Amazon with a release date of March 25.