Is reducing the resolution a good idea?

In brief

The size of photosites depends on their number and the size of the sensor. This is a physical characteristic of the sensor.

The resolution of the photo is, for its part, obtained using a software setting on the image processing chip. Changing this setting doesn’t change the physical characteristics of the sensor and therefore has no influence on the quality obtained.

In practice, the perceived quality depends on observation size (screen, 10x15 print, poster etc.) and can therefore vary with resolution. We do nevertheless still advise you to take your photos at the native resolution of your camera's sensor and to use your computer later for any modifications.

Reducing the resolution

A sensor, CCD or CMOS, has a given number of photosites — 10.6 million for example for the 10 Mpx CMOS sensor that's all the rage with some manufacturers. The image you get has one pixel per photosite and is slightly resized to give the camera’s native resolution: 9.98 Mpx for example for cameras with this sensor.

All cameras have a picture resolution setting however, that allows you to resize images at 7.5 or 3 Mpx for example. Sometimes you may be tempted to use this setting to counterbalance the excess of pixels on modern sensors (8 Mpx is enough for a good quality 30x40cm print) that sometimes cause high digital noise and reduce the dynamic range.


What’s happens when you do this? This very much depends on the camera’s image processing chip. Some resize it, which is to say they cut out the image and squeeze the area photographed; this is what happens with Extra Zooms and other Intelligent zooms used to simulate long focal length. Most of the time however they use interpolation: the image is shrunk to give the desired resolution, just as you can do with the Photoshop Image size feature or the GIMP Scale image feature.

This is a software feature built into the image processing chip and not a modification of the sensor itself: you can’t physically change the number of pixels on the sensor.

Noise reduction

Theoretically interpolation redistributes the light measurement errors (source of electronic noise) of several pixels and therefore reduces visible noise. Let’s take a look:


These images are "full size" extracts from photos taken with a Ricoh CX3 compact at 800 ISO. On the left, at the sensor’s native resoltuion (10 Mpx); in the middle at 6 Mpx; on the right at 3 Mpx. Note that the most pronounced billowing is on the first image: reducing the resolution seems to work here then, without nevertheless turning a grainy image into a clean one.

In practice?

The level of noise in an image doesn’t actually mean that much however. Depending on the visualisation resolution, the noise can become invisible to the naked eye. You therefore need to compare under identical visualisation conditions: same resolution (if viewing on screen) or same size print.


Here are the three images resized for 20x27 cm prints at 300 dpi. Can you see the difference? This time, the 10 Mpx photo is the best: with just a little more noise (in the orange on the right in particular), it is more precise, more accurate and the title Les révolutions is legible, while on the 5 and 3 Mpx images, it's blurred.


For small prints or visualisation on-screen, you don’t need too many megapixels: a Full HD screen is only 2 megapixels. Above, for a 10x13 cm print, the difference between the 10 Mpx (left) and 3 Mpx (right) images is subtle, not to say negligeable.

If this makes up your sole usage, you might be tempted to reduce your camera’s resolution so as to get lighter images: a 3 Mpx image weighs 1 MB, as against 3 MB for a 10 Mpx image and 5 MB for the forthcoming 14 Mpx cameras. On a 4 GB memory card, you can stock 800 14 Mpx images, but 4000 3 Mpx images…

Question of algorithms

There is however one last element you need to take into account: the algorithm used to reduce resolution, or in other words the way the colour of new pixels is calculated.


On the left, the image obtained with the CX3 set at 3 Mpx. In the middle, a 3 Mpx image created in GIMP from the original 10 Mpx image, without interpolation (the new pixel takes on the colour of the pixel that was closest to it on the original image): noise is heightened and we don’t advise this. On the right, same operation, but with Sinc interpolation.

Note how some characters, 'e' and 'g' in particular, are more legible: the algorithm used internally by the camera doesn’t do as good a job, which makes sense: the processing power of the camera’s processor isn’t anywhere near that of the computer, which means it uses much lighter software.

The conclusion here is categorical: even when what you’re using your camera for doesn’t require you to store images at high res and even if you’re looking to gain storage space, you’re better off taking photos at your camera’s native resolution and then resizing them on your computer and using your computer software settings (Sinc for GIMP, bicubic for Photoshop…)

And, if by some chance, you manage a perfect photo, you’ll be able to maximise your pixels and print yourself out a poster.