JPEG Quality in Photoshop

Regarding the information posted in this article from petapixel.com I decided to do a quick comparison to see if there is some truth to the statements mentioned in there.


I wanted to see what the perceived quality and file size is among various JPEG quality levels when exporting images from Photoshop CS5. I know the article states that sometimes quality is affected opposite from what you might expect but I figured that it was worth a shot. There are no absolutes with many things in life and this shouldn't be any exception.


I took two basic photos. I'm using a Nikon D700, 50mm f/1.8 AI-s lens, and SB-600 flash mounted on camera in TTL. The photos were taken indoors at f/5.6 and 1/60th. The RAW came in at 2832x4256 and 15.04MB. I didn't perform any post-processing and right from Bridge I simply exported 5 JPEGs at quality levels 6, 7, 8, 10, and 12.


Here is a screenshot for Levels 6-8 with one area of detail zoomed in at 200%:

Under close inspection from my end I can actually say that the artifacts that you normally see with JPEG compression is noticeably more going from Level 6 to Level 7 and notice that the file size is nearly the same. The difference is very slight but if you look hard enough (too hard really) you can find them. Level 8 is noticeably larger, as expected, and the amount of artifacts is less.


Here's a screenshot for Levels 10 and 12 with the area of detail zoomed in at 200% again:

Artifacts? There are a few in Level 10 and I would say they are less than the previous batch. What about Level 12? I can actually see a few less again. File size? more than doubled between 10 and 12. As you can see, we are getting quite large as well.


So, what's the quick verdict on this? There seems to be some truth to the statement that when you go from Level 6 to Level 7 you may lose some quality while the file size stays about the same. So Level 6 may be the better alternative to Level 7. However if you are counting pixels perhaps bumping to Level 8 will ease you, that is unless you are trying to stay within a specific threshold for file sizes (please be conscience of mobile users!)


As far as the statements about Levels 11 and 12 being "experimental" and not really adding to overall quality. Well I can't say there is obvious evidence of this because I did notice fewer artifacts when using Level 12 vs. Level 10. The file size did more than double.


Now I know this was not very scientific and there are many types of images and variables that should come into play regarding this, however, doing a thorough scientific evaluation wasn't the point of this little experiment. All I wanted to do is find out if these statements were completely bogus or if they actually had some truth to them. A good place to start with some further research is with the Adobe Photoshop development team. They can answer the differences in the algorithms being used among all the various levels...

Stitching

Two frames, stitched vertically

I've begun experimenting with stitching. That is, taking one xor more photographs (yes, xor, not or. one or the other but not both :) and stitching them together to obtain either a larger field of view and larger image. I like stitching photos because it gives a squarish crop, rather than the standard 3:2 for most cameras today. 3:2 works for many compositions but there are other times that I do prefer something closer to 1:1. I also like stitching because the end result is an image that is far larger and contains far more pixels than my camera can capture alone. This is what I want when I want to print large images and retain as much detail as possible. I'm not interested in switching to medium format film to obtain this same aspect ratio. I like working with digital and all that it provides. I'm also not prepared to dump the cost of a new luxury car entirely on a digital medium format camera as well. There are other options though...


One option is to clunk down about $2000 USD for a Nikon 24mm PC-E lens and use the shift functionality to create multiple images that may be easily stitched. Here you can shift up/down or right/left with the camera in either landscape or portrait orientation to satisfy the need for just about any composition. The other option to consider is a rail or pano(ramic) setup to allow me to create multiple images that may be stitched. This is the less expensive option and I can use it with any lens that I have. I decided to try a simple sliding rail setup and purchased the Really Right Stuff 192-PPP (Precision Plus Package). I didn't get a full panoramic setup for multi-axis rotation of the camera and lens because I didn't really want the added complication (and cost) and I could forgo some of that capability for now. I can always add more hardware later.


Stitching is now very easy when using the PS CS5 Photomerge feature. Results are had in just a few clicks and images are seamlessly blended and distortion free. There's not much to worry about when your images do not suffer from any perspective or distortion problems from the beginning. In particular, you have to avoid parallax distortion or errors between you images. To do this you have to rotate your camera and lens around then entrance pupil or no-parallax point. This is essentially what panoramic setups like the 192-PPP allow you to do.

Not all compositions and images will suffer from this optical phenomenon enough to prevent stitching. Much depends on the subject as well as the the perspective from where the image was taken from frame to frame. In fact, I was successful in stitching together several new images recently without the aid of the 192-PPP. The image to the right is composed of three individual images stitched together. The combination of using a long lens (400mm) and using subjects that were some distance away reduced the amount of vertical movement I made to compose each frame. This reduced the amount of distortion from frame to frame even though I was not rotating or moving the camera and lens around the no-parallax point. I have to owe all of the credit to the awesome power of the Photomerge tool though.


In the end, you'll most likely have to contend with these distortions in your some of your compositions and there is no substitute for knowing how and having the correct tools to do so. You can't always rely on the Photomerge tool to bail you out.


I placed my camera and a 25mm lens with the 192-PPP all on the tripod and then setup a scene to determine the no-parallax point for this lens. I've read that people like to call this the nodal point but from what I understand, this term is technically incorrect for what we are trying to accomplish here. I'll just stick with no-parallax and move on. The scene I setup was just a simple lineup of a near and a far object. Through trial and error I determined the approximate no-parallax point for this lens.


There are a couple of pointers or tips that I found which may be very helpful:

1. Make sure the camera is level
2. When rotating the camera to the left or the right, if the far object moves in the same direction in relation to the near object, then you are too far forward, move back
3. When rotating the camera to the left or the right, if the far object moves in the opposite direction in relation to the near object, then you are too far back, move forward

I then proceeded to go through all of my lenses and do the same. With prime lenses, you just have to do it once. With zoom lenses, you have to do it for each focal length you believe you will use. For example, with my 80-200mm, I determined the no-parallax point for the 80, 105, 135, and 200mm focal lengths. Once that was completed I simply printed these out and made a handy little cheat sheet that I carry with me in my bag that I can refer to each time I need to know the no-parallax point for one of my lenses. Now I can make easily stitch-able images using the no-parallax point without having to calibrate it each time.