Noise – a case study

Other pictures taken with the Samsung T929 (Memoir) are available on my Flickr account.

So I decided to do a little experiment: take 3 pictures with 3 devices (one each) at night in the same location around the same time! The goal? Evaluate how megapixels affect low-light performance.

The megapixel myth tells us that more megapixels is not necessarily better. It generally means smaller pixels, which means more heat, which results in more noise, which impacts low-light performance…

Of course there are many variables here: differences in optics, sensor type/size, image-processing/noise-reduction software, and more. Because this is meant to be a real-life test with real-life devices, I can’t control these variables – I can only pick suitable devices.

Ideal devices:
Samsung T929 (Memoir), 8 MP
Nokia N82, 5 MP
BlackBerry 9530 (Storm), 3 MP

Unfortunately I no longer have access to the N82 or the Storm.

Final devices:
– Samsung T929 (Memoir), 8 MP
Nokia N85, 5 MP
Nokia 5800 XpressMusic, 3 MP

I’d planned to use my T-Mobile G1 instead of the 5800, but it’s been raining a lot in San Francisco lately, so I’d been waiting for a clear night to do this experiment, and when the time finally came, my G1 needed charging :)

After taking the pictures, I decided to edit them and match brightness, color-balance, and size (scaling them all to 3 megapixels). Here is the result – first with the 5800, second with the N85, and third with the Memoir (click to enlarge):



Here are some closeups – 5800 to the left, N85 in the middle, and Memoir to the right (click to enlarge):


And the winner is? The 8 megapixel Samsung T929 (Memoir) – even scaled to 3 megapixels, there’s more detail and less noise in the picture…

But wait, this reinforces the megapixel myth! Yes, but the 5 megapixel Nokia N85 and the 3 megapixel Nokia 5800 both generate a lot of noise to begin with. And perhaps the Memoir uses a larger sensor (i.e. all 3 devices use similar sized pixels), spreading the noise over more pixels?

Discuss :)

Advertisements

8 Responses to Noise – a case study

  1. abubasim says:

    You should’ve included the i-mobile 902 – as a reference.

  2. Tom says:

    But… isn’t the comparison damaged by resizing the pictures?

    Noise appears on the image as a per-pixel anomaly, rather than an anomaly that is scaled larger with more pixels. By resizing the higher resolution pictures, the noise will be averaged out by the resampling algorithm, giving a smoother appearance with the /impression/ of less noise.

    Could you publish the original images?

  3. whatleydude says:

    Like it…

    I got some pics with my N86 if you want ’em? ;)

  4. tnkgrl says:

    @abubasim I no longer have access to the i-mobile 902.

    @Tom I was debating whether to upsample the pictures to 8 MP or downsample them to 3 PM. Ultimately I decided to downsample them because that more realistically reflects what people are likely to do (to email them or whatnot). As for posting 8 PM versions, I don’t remember exactly what kind of brightness/color adjustments I performed to each original, unfortunately. Maybe I can try this experiment again using the G1 and adding a regular point and shoot.

    @whatleydude Can you post them on flickr?

  5. Tom says:

    I’d have thought to compare the noise levels, you’d not want to resample them differently at all? Just compare them all au naturel, with or without equal simple scaling to allow observation of individual pixels?

    Interesting topic, though, and top marks for pursuing it scientifically – I’m sticking with my trusty N95 for the time being :)

  6. tnkgrl says:

    @Tom I don’t think my experiment is very scientific, but it’s interesting :)

    As for the Samsung T929 (Memoir), I’m very impressed! I think the camera is better than that of the N95. Wait until you see my camera review…

  7. KlipschFan says:

    It looks like the N85 has some auto focus issue with low light when compared to the other two.

    I think I remember Alex Lindsey saying something about noise coming mostly from the blue part of the spectrum.

    Since I am YourAverageIdiot when it come to photography, I’m not exactly sure how to interpret his statement. I did notice that the 8500 and the Memoir color of the dark sky were comparable in the first set. But then again…”I decided to edit them and match brightness, color-balance, and size…”

    I love the question, but I think their are too many variables to definitively answer the question. Still, I say hats off to the person who tries.

  8. adaptg says:

    Better lens on the Nokias for sure. Look at the colour fringing effects of the light sources, even the moon.

Leave a Reply

Please log in using one of these methods to post your comment:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s

%d bloggers like this: