Digital Photo Processes

-- Introduction -- Lens & image -- A/D conversion -- Lumix & Leica --  Sharpness -- CCD noise1 -- CCD noise2 -- 5 years later -- 

1 - CCD photo sensor noise

The weak link in a compact digital camera in terms of image quality is usually the photo sensor. Its electronic circuits add noise to the voltage that has been generated from the accumulation of photons in the individual pixel elements of the sensor. The conversion of electrical charge into voltage is associated with some noise, as is the amplification of this voltage to obtain higher ISO settings and thus higher light sensitivity. The photon conversion and addition of noise is shown schematically in the block diagram.

Figure 1 Light conversion to electrical voltage.  The output voltage is digitized, then processed for RAW and JPG data file storage inside the camera.

Not only is electrical noise added, but the photon flux itself has a noise component. Its magnitude is the square root of the total number of photons. The photon noise is not an issue for small sensors, because it is swamped by the ectrical noise. In a DSLR with a large sensor and at low ISO settings it becomes the limiting factor for the maximum signal to noise ratio of the camera. The table gives a numerical example of how the output signal-to-noise is affected by the amount of light received and the ISO setting. It assumes a constant lens aperture opening. Only at ISO 100 is the output S/N determined by the incoming photon number. Its noise is higher than the combined electrical noise. At higher ISO settings the two electrical noise sources dominate.

One can get some idea of the camera's noise by looking at the image that is generated when no light hits the sensor. Merely expose with the lens cap on or with an opaque hood over the protruding lens, if no lens cap is available. The pictures show such dark noise for a Lumix LX2 at ISO 800 and in RAW mode. The file was processed with Adobe Camera Raw in Photoshop Elements 6. Sharpening and noise reduction were set to zero. The Exposure slider was set to 2. 

Blotches and streaky patterns are noticeable. The noise increases with exposure time, but slowly as the 100:1 exposure time difference shows.

2 - Camera dark noise tests

This type of dark noise analysis was performed on four related cameras, Lumix LX2 and FZ50, and Leica D-Lux 3 and V-Lux1. The ISO 800 and RAW mode were chosen to show maximum noise. Exposure time was 1 second. Noise is shown at 100% and 200% magnification and with the color removed. 

From this qualitative dark noise analysis it would seem that the D-Lux 3 has the lowest noise. The FZ50 and V-Lux 1 are about equal but with the FZ50 having a slight edge. Note that the two different camera sensor types have different noise patterns. Overall the LX2 may be the worst performer because of a less random noise pattern. Furthermore the noise patterns are not locally fixed in the image pixel matrix which some further testing confirmed. They are caused by time varying noise.

3 - Scene Tests

I tried to put a scene together that would show how the camera noise affects the image quality. This is not that easy. Below are cutouts at 100% from the actual photos which were found to be representative. All were taken at ISO 800, F:2.8, 1/40 s, RAW mode and processed with ACR default settings. Notice how the removal of color brings out the noise even more strongly in the black and white photos.

The qualitative conclusions from these photos are similar to the ones from the dark noise tests above.

It would be very tedious and time consuming to create scene tests for different ISO settings and ambient light input levels to evaluate the image quality of the four cameras under changed conditions. Therefore a quantitative test was sought that would yield numerical results that could be directly compared between cameras. As it turned out this test is not that difficult, provided that one can provide a reproducible and color neutral light input to the camera lens so that there is a stable light level reference. 

Continued on page 2 

... where you will find a lot of graphs and numbers which may look intimidating. Basically they lead to conclusions which are summarized on the Lumix & Leica page, so you can skip page 2 and save me some bandwidth. On the other hand, the test technique that I show here is easily duplicated and allows meaningful comparisons between any number of digital cameras as to their absolute and relative noise performance in RAW and JPG modes, at least to a first order, though there can be differences in how smartly the noise reduction algorithms are set up. Those differences are not likely to show up in this analysis.

-- Introduction -- Lens & image -- A/D conversion -- Lumix & Leica --  Sharpness -- CCD noise1 -- CCD noise2 -- 5 years later --