Canon Bows Pixel-Shift; Finicky, but Delivers

Canon’s recent R5 firmware update added a few features and fixes, the primary headline grabber being the introduction of Canon’s first pixel-shift high resolution capture feature. Pixel-shift effectively multiplies the resolution of the camera sensor by about 4 times by taking 9 pictures from very slightly shifted perspectives (changing perspective by less than a pixel) and then combining the them into a single image, interlacing the pixels as though they were taken at the same time.

The feature works best when the subject is perfectly still and the camera is solidly locked down.

Screenshot of Extreme Crop

Most other major brands already had this feature in their cameras, but the practicality of those systems varied a great deal. For instance, Panasonic’s version of the technology is quit intelligent, allowing slight movement in the subject and being useful even hand-held. Sony’s, on the other hand, requires so much precision that even tripod-mounted attempts often fail.

Early reviews of the Canon version have been mixed to negative. While it does not frequently fail to work, the resulting image is available only in JPEG, and the algorithm used to interlace the images requires fairly perfect alignment. Without that solid stabilization and lack of subject motion, the result includes posterized parts that look much worse than a low-res image.

Most tests used for comment on camera forums have been using simple shots taken in home offices, which is usually a poor way to test the benefits of the extra high resolution. Below we include three images comparing a ludicrously-cropped portion of an image of a brittany spaniel figurine. The first is the cropped image straight out of camera. The second is the pixel-shifted version. For comparison, the third is the normal version of the photo post-processed in Adobe Lightroom with its “Enhance” feature, that uses software to double the resolution.

Cropped But Unmodified

The photos were taken with the extremely sharp Sigma 28mm f/1.4 Art to minimize the loss of detail due to the resolving capability of the lens. 28 mmm was chosen as the focal length to require extreme cropping in order to maximize the perceived effect of any extra resolution. The figuring was seven feet from the camera, and the aperture was set to f/4 to maximize resolution. As can be seen in the image above, the crop (the lighter rectangle in the middle) is about 1/1000th of the original image.

Canon Pixel-Shift

This image (right) shows the crop straight out of camera.

The image at left shows the equivalent crop from the pixel-shift version, nominally at four times the resolution.

The second, lower image at right shows the equivalent crop of the unmodified image after being put through Adobe Lightroom’s “Enhance” feature, which attempts to double the resolution using deep learning software. The pixel-shift image, by contrast has two times the nominal resolution as the Lightroom-enhanced image.

Lightroom “Enhanced”

The pixel-shift image weighs almost 200 MB, about four times the size of an R5 RAW file – this despite the fact that it saves as JPEG format. Not surprisingly, Lightroom on a 2017 iMac opens and edits the image roughly four times more slowly than the typical R5 file. This is doable with current versions of Lightroom Classic, especially when dealing with post-crop files. Processes such as creating 1:1 previews do take a disruptively-long time to generate.

Extreme resolution junkies – or just plain masochists – will be disappointed to learn that the pixel-shift images are too large in resolution for Adobe Lightroom to apply its Enhance feature on them. A work around for this is to save a crop as a separate file; load that back into the library, and then apply Enhance. Doing so will generally improve the perceived resolution slightly. Interestingly, it seemed to resolve some of the small artifacts sometimes present in the pixel-shifted images. An image showing that pixel-shift+enhanced effect isn’t included in this review because the differences are not significant enough to be visible in the pictures downsaved to appear on a web page.