Thursday, 28 April 2011

Bokeh comparison @ 14mm and 20mm

Many people are looking for camera systems that can give a thin depth of focus (DOF). With a thin depth of focus, objects that are beyond the focus distance, or closer, are out of focus.

The Micro Four Thirds system is not ideal for getting thin DOF. To get a thin DOF, you are better off buying a camera with a large sensor, for example full frame DSLR cameras.

However, it is still possible to get a thin DOF with Micro Four Thirds if you use a close focusing distance. I have evaluated the out of focus rendering (bokeh) at close focus using three lenses: The Lumix G 14mm f/2.5 pancake, the Lumix G 20mm f/1.7 pancake and the Lumix G 14-42mm f/3.5-5.6 kit zoom lens.

I took the same picture using the three lenses at various apertures. Here are the full images at maximum aperture:



Lumix G 14mm @ f/2.5
Lumix G 20mm @ f/1.7


Lumix G 14-42mm @ 14mm f/3.5
Lumix G 14-42mm @ 20mm f/4.1

The focus was set on the emblem on the bell in the middle left part of the image. I used the Panasonic GH2 at base ISO, and the shutter speed was around 1-6 seconds. I used a tripod, and also two second shutter delay, to avoid camera shake.

The images above are taken using the maximum aperture available with the given lens. Hence, the DOF is as thin as possible, given the focal length and focus distance.

To better evaluate the bokeh, I have made 100% crops from two parts of the image (click to enlarge):



The first crops are from the focus area. From these images, it could look like the 14mm pancake lens is unsharp. However, these images were taken primarily to evaluate the bokeh, not the sharpness, and the focus point might be slightly different between the lenses. In my experience, the sharpness of the 14mm pancake lens is rather good.

In the seconds image, we see the out of focus highlights. I suppose one could say that neither of the lenses give a very nice bokeh. They have various problems. They all exhibit some ringing, but it seems to be worst at 14mm. Also, the bokeh is uneven, and "dirty", "swirly".

The 20mm pancake lens shows the most non-circular highlights, both wide open and closed down.

The 14-42mm lens shows some strange irregularity at f/5.6, at both 14mm and 20mm. This could look like a construction error of the aperture diaphragm. However, it is not likely to pose much of a problem, since only at very close focus distance would you see much out of focus rendering at f/5.6

In terms of roundness, the 14mm pancake has the most consistent appearance.

Tuesday, 12 April 2011

GH1 and dead pixels

Back in the day when people started replacing CRT computer monitors with LCD panels, dead pixels was a big deal. When you received the LCD panel you had ordered, you would have to review it to find the number of dead pixels, and consider whether or not to return it. Some claimed that dead pixels could be massaged with your fingertip and revived. After the production process improved, though, dead pixels has ceased to be a problem with computer monitors.

A computer monitor with 1280x1024 pixels has a total of 1280x1024x3 individual dots, one for each primary colour. This makes almost four million individual dots that make up the image you see. Previously, I took a macro closeup picture of a computer monitor to reveal the pattern of red, green and blue dots.

With camera sensors, the number of pixels is counted as the number of individual dots, each capable of seeing only one of the primary colour. Generally, these pixels are arranged in a Bayer-pattern, with two green pixels for each red and blue:


So a camera with 12 megapixels has 12 million individual light measuring cells, three times as many as the dots on a typical computer monitor. So is dead pixels a problem with cameras?

I tested my Panasonic GH1 camera. I took one very underexposed picture (which turned out black) and one very overexposed picture (which became white). The JPEG images are here, straight from the camera:



You don't need to look at them, though. Trust me, they are completely even, with no signs of dead pixels whatsoever.

However, what if the camera corrects the dead pixels in the JPEG files it generates? Perhaps there are "holes" in the image data, which is filled in by the JPEG engine in the camera's algorithm.

To check this, I opened the RAW files in a third party RAW converter program, the UFRAW. This did in fact reveal some dead pixels. I found eight pink dots. The colour pink is due to the green sensor element being dead, I suppose.

In the image below, a pink ring has been put around the dead pink pixels. After scaling the image down to 1000 pixels wide, the dead pixels themselves are of course not easy to spot anymore.


When investigating the black image in the RAW converter, I found no evidence of stuck pixels. So no pixels were generating a "phantom" light even though the exposure was non-existent.

I looked at other exposures, and found the dead, pink pixels in the same spots. So these pixels are definitively permanently dead on my GH1 camera.

Is this a problem? Hardly. As we saw, the JPEG engine is clever enough to mask these dead pixels, so when using JPEG images out of the camera, don't worry.

When using the RAW images, though, the dead pixels might disturb the image when using high resolution prints, for example. I would guess that the supplied RAW conversion program fixes these problems automatically, so this is likely only to be a potential problem with third party converters. And even then, eight dead pixels out of a total of 12 million is not exactly a huge percentage. It is very unlikely that this will cause any unwanted side-effects.

Appendix

Panasonic G series cameras do have a function to map these dead pixels. To do so, use the function "Pixel Refresh", which can be found in the custom menu (on the last page). The custom menu is the one with the "C" and wrench icon.

Saturday, 9 April 2011

Self portrait on a bicycle

The Lumix G 8mm f/3.5 fisheye lens is fun to use, due to it's extremely wide perspective. This type of lenses are commonly used in extreme sports videos, like skateboard and BMX. I don't do this type of things myself, but I figured I could use the lens to record myself bicycling anyway.

To do this, I attached the Panasonic GH2 camera to the front fork of the bicycle. I used a Manfrotto Super Clamp, which is essentially a clamp with a tripod head attachment stud. To be able to attach the camera, I also used a tripod ball head, and I chose the Benro BH0, which is pretty compact. After putting it on my bicycle, it looks like this:


Here's the video I recorded:



I prefocused at about 50cm, and set the camera to Manual Focus (MF) before starting the video recording. This was done to avoid focus hunting during the video. The 8mm fisheye lens focuses very quickly, but still, I wanted to avoid short periods of out of focus footage.

I think the colours look a bit dull in this video. It was a somewhat dull day, but still, I think that the next time, I will increase the colour saturation setting before using this lens for videos.



Sunday, 3 April 2011

Hacked GH1 vs GH2: Video quality comparison

A matter of some dispute is which camera gives the best video quality: The hacked GH1 or the newer GH2.

The GH1 has earned some popularity due to the possibility to change the firmware. Adjusting the firmware is generally referred to as "hacking" the camera. There are many options, and the most useful ones are the possibility to get native 1080p 25fps (with the PAL version), and to increase the bitrate. I have done both with my GH1.

I've compared them before, and my conclusion back then was that the GH2 had better auto white balance (AWB) for indoor lightning, but other than that, it was hard to find much evidence of better video quality. I was advised that I should try to pan the cameras while recording foliage.

There are two problems with that: One is that there is no foliage in Scandinavia at this time, since spring is not yet here. The other is: How would I know that I have panned at the same speed, giving comparable footage?

For the second problem, Technic LEGO again comes to the rescue. Previously, I made a rotating object using Technic LEGO, and video recorded that. This time, I made a platform for the camera out of Technic LEGO, and used that to pan the cameras at the same speed. Here is how the platform works:



Putting the platform outdoors enabled me to record the same footage using both cameras. I used the same image parameters for both cameras: ISO 200, f/3.2, 1/400s shutter speed, outdoor sunny white balance, manual focus. Due to limitations in the cameras, I could not use the same frame per second count. I used 1080p for both, but 25fps for the GH1 and 24fps for the GH2.

I used the Lumix G 14mm f/2.5 lens, which is more than sharp enough for this test.

In real life use, one would not normally use a shutter speed of 1/400s. The normal speed to use is twice that of the framerate, which is 1/50s in my case. This is called a 180° shutter, since it is open half the time on average.

One reason why a 180° shutter is normally used, is to get motion blurring for objects moving across the frame. Without motion blurring, the movement can look unnatural on film, since the object will appear as if it materializes in different spots at different times. This is confusing for the person watching. I have a discussion about this, and the need for Neutral Density (ND) filters here.

In my case, though, I want to avoid motion blurring. The whole point of the experiment is to see which camera resolves the most details, and then I must make sure that the video stream contains as much details as possible to begin with. Motion blurring typically makes the footage softer.

Due to the hacked GH1 having a higher bitrate, it generated larger video files. The GH1 gave 5.4 MB/s, while the GH2 gave 2.8 MB/s.

Here are the two video recordings:



GH1



GH2

It's somewhat difficult to evaluate the quality of the video by looking at the videos, especially since YouTube compresses them anyway. So to aid in comparing them, I have grabbed single frames from them to compare.

Here are two similar frames (click for larger versions of them):


GH1


GH2

The first thing we can see, is that the GH1 still image is brighter, despite having the same image parameters. I interpret this to mean that the ISO scales of the cameras are not the same: The GH1 ISO scale corresponds to higher ISO values with the GH2. It is a well known and documented fact that the GH1 ISO scale is generous, meaning that a given ISO value gives more sensitivity than the same value for comparable cameras. The GH2 ISO scale is more normalized.

Another thing we can note, is that the trees lean a bit to the right. This is due to panning the camera, and the rolling shutter. I've evaluated the rolling shutter effects of the two cameras before, and found them to be comparable. Rolling shutter can create artifacts when using both the GH1 and GH2, but for normal use, it is not a real problem.

Here are direct comparisons between two areas from the two video streams. They are shown here in 100%, i.e., not rescaled and not sharpened.


Since the exposure is slightly different, I have also done an autolevels on the images, to make them comparable. Here they are:


Conclusion

One could be tempted to think that the GH1 gives better video quality, due to the twice as high video bitrate. But the GH2 has other advances. We don't know for sure just how the GH2 handles the video compared with the GH1, but it is reasonable to guess that it samples more pixels as a basis for the video output, and that the compression algorithm is better.

So, which video stream is better in my test? I think they are quite similar. Perhaps one can conclude that the GH2 gives somewhat better contrast and sharpness. The GH1 has slightly washed out colours, I think. But the difference is small.

My conclusion so far is that the GH2, even with lower bitrate, gives slightly better video quality. But both cameras are very competent.