Tuesday, October 13, 2009
Noise: Lose It, Part II
Reducing chrominance and luminance noise at capture
Noise comes in three types or patterns:
1) Random noise 2) Fixed-pattern noise 3) Banding noise
Noise often has two components—brightness and color:
4) Image noise 5) Luminance noise 6) Chrominance noise
Knowing the type and kind of noise produced will help guide you to solutions to reduce it. There are three types of noise: random noise, fixed-pattern noise and banding noise.
Random noise appears as both luminance (light and dark) and chrominance (hue/saturation) variations not native to an image, but produced by the electrical operation of a capture device. The electrical signal produced in response to photons is commingled with electrical variations in the operation of the capture device. Random noise patterns always change, even if exposure conditions are identical. Random noise is most sensitive to ISO setting. Again, digital cameras have one native ISO setting; higher ISO settings artificially boost the signal produced by the sensor and the noise accompanying it. The results? You get a brighter picture from less light and exaggerated noise. Since the pattern is random, it’s challenging to separate the noise from the image, especially texture, and even the best software used to reduce it through blurring may compromise image sharpness; how much depends on the level of reduction.
Fixed-pattern noise (“hot pixels”) is a consistent pattern specific to an individual sensor. Fixed-pattern noise becomes more pronounced with longer exposures. Higher temperatures also intensify it. Since the pattern is consistent, it easily can be mapped and reduced or eliminated.
Banding noise is introduced when the camera reads the data produced by the sensor; it’s camera-dependent. Banding noise is most visible at high ISOs, in shadows and when an image has been dramatically brightened. This type of noise is obvious and objectionable; the regular row and column patterns from the sensor quickly call attention to the presence of banding noise, and it’s challenging to reduce without severely compromising image sharpness.
Noise can be broken down into two classifications: chromatic (hue/saturation variances) and luminance (brightness variances).
Chromatic noise produces a more “unnatural” appearance, and it’s easier to reduce without compromising image sharpness than luminance noise. Chromatic blurring is less noticeable than luminance blurring because human perception tends to see color differentiated within contours, even when that differentiation isn’t precisely true. It’s a convenient optical illusion. Larger chromatic variances may result from Bayer pattern demosaicing. Digital sensors typically capture photons with a repeating array of two green, one red and one blue photosite(s) that register separate luminance values for each site. This data is then processed, “averaged,” if you will, to generate a final color, such as brown or lavender, or even a specific green, red or blue. If done under suboptimal conditions, such as underexposure, larger areas of color variance may occur and will require additional postprocessing. Extreme amounts of chromatic-noise reduction may result in reduced saturation, especially along contours separating strongly contrasting colors.
Page 1 of 2