Doc Russ' Magic Seven Step Elixer for Image Processing
Step 3. Remove or reduce noise.Image noise must be removed before contrast expansion or detail enhancement, or it will be increased by those steps.
Step 3a. For dropout or random noise apply the Fovea Pro Filter > IP*Rank > Hybrid Median routine or the ClearID > Universal Noise Remover (select the Hybrid Median function) and adjust the neighborhood radius slider to eliminate the fine-grained noise. (This is superior in several ways to the Photoshop Filter > Noise > Median routine: 1) it handles color properly, rather than just operating on brightness; 2) it is more isotropic, rather than favoring horizontal and vertical directions over diagonal ones; and most important 3) it preserves fine lines, spaces and corners that are erased, filled or rounded by the median routine.) For digital camera images in which the random speckle noise in each channel is independent and different in magnitude (e.g., the blue channel is often noisier), use Photoshop's Filter > Noise > Reduce Noise routine in Advanced mode, and work on individual channels. (Note - this function is not fully documented and has a great many adjustments, so it takes a while to master.) The figure shows a comparison, using a digital camera image taken with a high ASA setting.
Tech details: Random or speckle noise increases with poor lighting (requiring higher gain in the electronics). Most camera chips also have at least a few dead or locked detectors that cause dropout (black or white) pixels. A common (but mistaken) approach to reducing it uses neighborhood averaging or Gaussian smoothing, which also reduce resolution without actually eliminating the problem. Rank-based filters compare pixel values to their neighbors and replace extreme pixels with a more likely value from nearby. The classic median filter reduces noise effectively but also removes lines, fills narrow gaps, and rounds corners. The hybrid median alleviates these problems, at the cost of increased processing time (rarely a problem with modern computers). The particular version used here also works correctly in color space by using vectors, whereas the simple median just ranks the pixel brightness values.
Step 3b. For pattern noise, apply the ClearID > Pattern Remover routine and adjust the sliders to remove the noise pattern as shown in the example (a half-tone printed image). The dialog shows the adjustments: the Low Frequency Cutoff controls the size of the central blue region on the Fourier transform, which preserves the overall brightness and contrast of the image; the Frequency Acceptance controls the number of spikes in the transform that are marked in red for removal; the Notch Radius controls the size of the red mask areas.
Tech details: Any periodic pattern in the image, whether from printing technology, electronic interference, or vibration, can be isolated as a set of 'spikes' in the Fourier transform. Eliminating the specific frequencies and orientations marked by the spikes removes the periodic pattern without altering any of the other information that is present. The Pattern Remover dialog shows the Fourier transform and makes generating the filter or mask (the two terms are often used interchangeably) that removes the spikes very straightforward. It is not necessary to understand the math behind the Fourier transform, just to observe the presence of spikes and the effect of the mask settings on the resulting image.
The Seven Steps:
- Back to course listing
- Back to the beginning
- Step 0. Average multiple frames.
- Step 1. Adjust the colors.
- Step 2. Remove or reduce artifacts.
- Step 3. Remove or reduce noise.
- Step 4. Remove or reduce blur.
- Step 5. Maximize image contrast.
- Step 6. Enhance sharpness and detail.
- Step 7. Save results and document.