Friday, December 19, 2008

Are RAW images really RAW

I have seen that a lot of arguments in various forum regarding RAW images from Nikon and Canon especially concerning Noise Reduction(NR). There are accusations that the other brand plays unfair because they change the raw data where NR is applied to the RAW image.

This argument occurs because most of them are not aware how the image is acquired. I will attempt to describe how an image is acquired by a camera in as simple manner as possible but this is a crude way to describe and not essentially scientific.

Today CMOS (Complementary metal–oxide–semiconductor) sensors work in the basic principle that when light falls on the sensor pixel unit, the energy from the light is converted to voltage. The transistor attached to the sensor pixel units amplify this signal. The sensor has a Bayer Filter which is a color filter. The filter allows only light from one color, red, blue or green. Usually green is represented twice that of blue and red. So the sensor captures only 50% of incident green light, 25% incident red and 25% incident blue light. This signal is converted to voltage, amplified and demosaiced into an image. This demosaicing is the process that happens in all cameras and the image is formed after demosaicing. This is both hardware and software process. The resultant signal is the RAW image. Thus there has to be some processing prior to getting a RAW image. The way different colours are rendered, shadow/highlight detail captured, signal to noise ratio are all affected by this hardware/software processing. All these happen at the level of the Sensor Complex. There is no way to obtain an image without this process. You will have only electric signal otherwise.

This RAW image can further be processed to JPEG or TIFF or any other format. To accurately state the RAW image is not unprocessed but less processed.

Why is it then possible to adjust varying image parameters like exposure, brightness, color, contrast etc better in a RAW image to a JPEG/TIFF image? The answer is in the fact that the Gamma or the luminance intensity of the RAW file is wider than JPEG. Let me explain. The image data is represented as light levels. In a 12 bit raw image there is approx. 4000 light levels whereas the 14 bit has approx. 16000 light levels. The JPEG has 256 light levels (8 Bit). So this huge information is available for manipulation.

So what does RAW have?

More Information so less posterization.

White balance is not yet encoded so this can be altered in a wide range,

Color coding from bayer filter can be done by the RAW converter ( like lightroom, aperture etc) which can be more powerful and your visual input provides more accurate adjustment of colors.

These are the reasons why RAW images are superior for manipulation purposes ( please note I said superior specifically if manipulation is important for the photographer).

Good but I never discussed Noise Reduction (NR) in this entire process. In fact if you have noticed NR by a RAW converter is usually inferior to the NR done by specialised software (like noise ninja).

To understand this we need to identify how noise is introduced in an image. In the image processing chain noise is introduced at the point where the electrical signal from each pixel unit is amplified by the transister. This is actually a very tiny signal (voltage) that needs to significantly amplified to obtain meaning information. Unless this stage of amplification is completely eliminated from the image processing chain there is always going to be noise. But rather than noise the most important parameter is Signal to Noise Ratio (SNR). This is least discussed by many photographers. In low light, (high ISO images) the amplification is more so noise is more than signal thats why the image is degraded.

So in camera NR occurs at 3 levels.

1. At the sensor pixel level, depends on sensitivityof the sensor pixel unit to light. The more sensitive the pixel unit, the less amplification of signal needed and less noise (more SNR) is present in the image.

2. At the demosiacing level where the electric signal from sensor complex (sensor pixel unit and transistor) is covnverted to image signal.

3. At the RAW conversion level ( where Jpegs are created in camera).

We cannot control the first two levels. This is determined by the manufacturer of the sensor in level 1 and camera manufacturer in level 2. ( This is the reason for different noise levels in different camera maufacturers, though the sensor might be the same).

We can manipulate the last level. Since SNR is more dependent on the semsor complex, RAW NR does not offer as significant advantage as compared to other forms of image manipulation.

If you were able to follow me you will realise why RAW images from different manufacturers appear to have different levels of noise reduction. If the manufacturer feels that their cameras have more resolution they can afford to be more aggressive with NR since ultimately all that matters is Signal to Noise Ratio (SNR) as this is what we finally see. unfortunately this is least discussed parameter. More on SNR in future blogs ....

No comments:

Post a Comment