MLUnsold Digital Imaging

Detect & Remove Residual Structured Noise after Calibration and Stacking

Calibration and stacking reduces pixel non-uniformity and random noise from dark current, shot noise, read noise, and camera off-chip noise since random noise does not build up as fast as signal from the objects in the image. But structured noise from electronic interference do build up as fast as signal from the objects and are not removed by calibration and stacking. Below is a method to detect if your camera has residual structured noise and a method to reduce or remove structured noise from an image.

Detecting Structured Noise

If you are using a DSLR then use the Image Set Operations | Digital Camera Raw Convert tool to convert CR2, NEF, or other raw bias frames to Bayer At Capture White Balance format. If you are using a one shot color or monochrome CCD then start with raw Bayer format bias or monochrome bias frames as they come from the camera.
Average combine 10 bias frame using the Image Set Operations | Combine Files/ HDR Add tool.
Open the average combination bias frame and one of the bias frames used in the combination then use Special Functions | Image Math to subtract the average combination bias from the single bias frame. Use the Split Luminance button on the left vertical toolbar to split the RGB image created by Image Math into a single channel gray scale image. Let's call the difference image the structured noise bias. Random noise is reduced in the structured noise bias from the subtraction of the average bias.
If you are using a DSLR or one shot color CCD then use the Color | Split Bayer CFA tool to separate red, green, and blue pixels.
Open Geometric | Image Size and press the Crop To Max 2^N button then press Apply to size the structured noise bias to a power of 2. This step reduces the effect of the image width and height on the frequency spectrum that is used to detect structured noise. If you are using a DSLR or one shot color CCD then apply this step to the separate red, green, and blue images created from the previous step.
Open the Smooth Sharpen | Frequency Filter and check the Show Spectrum box then apply to the power of 2 structured noise monochrome or red, green, and blue images.

The QSI 583 structured noise spectrum above has no structured noise which would show up as bright vertical or horizontal lines or spikes. Only the bright center DC pixel is shown which represents the average color of the image. This is what the structured noise spectrum looks like from a camera with little or no noise from electronic interference.

The structured noise spectrum of a Canon 60Da is shown above. The horizontal and vertical line through the center bright DC pixel is from periodic noise that is not removed by calibration and stacking. The amplitude of frequencies along the horizontal and vertical lines can be reduce using the method shown below and will slightly improve the image.

The structured noise spectrum above is from a SBIG ST-10 and shows several bright vertical lines on each side of the center DC pixel. These lines are from structured or periodic noise that is not removed by calibration and stacking. The Frequency Filter can be used to remove the frequencies marked by the vertical lines to reduce structured noise in the calibrated and stacked image as shown below.

Reducing Structured Noise

Use the Smooth Sharpen | Frequency Filter tool to check for structured noise in the calibrated, stacked, and stretched image.

The calibrated, stacked, and digital development stretched image is shown on the left. On the right is a duplicate of the image with the Frequency Filter applied with the Show Spectrum option. Note the fine horizontal and vertical lines that match the structured noise bias frame shown in the previous screen shot.

Use the Smooth Sharpen | Split Frequency tool to create a frequency spectrum that can be painted on to remove unwanted frequencies.

Open the Smooth Sharpen | Split Frequency tool. The default setting will create a frequency spectrum and an update image that is the modified version of the initial image using the modified spectrum. The Smooth Sharpen | Combine Frequency tool will be opened automatically and setup with the initial M1 image, spectrum, and updated M1 combination image as shown by the next screen shot.

M1 image is shown on the left. M1 updated image at the center and its initial spectrum image on the right. The Combine Spectrum tool is shown on the far right.

Open the Local Smooth, Sharpen, and Paint tool from the left vertical toolbar. Use the Horizontal Line Fill and Vertical Line Fill options with Intensity = 0 or black to paint out the fine vertical and horizontal lines on the spectrum as shown above. ***Only paint with 0 on the spectrum to eliminated the vertical and horizontal line frequencies.*** Press the Apply button on Combine Frequency to update the combination image using the painted spectrum.

Use the Zoom 400% button on the top horizontal toolbar to zoom the initial and combination M1 image. Check the top left corner of the combination image for ringing near the top and left side edge. Reduce the length of the black painted horizontal and vertical lines to remove ringing if present. Press the Apply button on Combine Frequency to update the combination image using the current painted spectrum. The undo toolbar button can be used to remove unwanted paint strokes.

Inspect other parts of the combination image with zoom 400%. The calibrated, stacked and digital development image shown on the left has a fine pattern of short vertical and horizontal lines in its background. The combination image on the right was built with the painted spectrum with structured noise removed and has no fine vertical and horizontal line patter. Removing the fine structured noise improves processing that follows.

On the left is the calibrated, stacked, and DDP stretched image with no structured noise removed and 10 iterations of adaptive Richardson-Lucy deconvolution applied. Deconvolution amplifies the fine vertical and horizontal structured noise line patterns. Removing structured noise reduces the vertical and horizontal line patterns from the stretched image and improves deconvolution sharpening as shown by the image on the right.

M1 image is shown on the left. M1 updated image at the center and its initial spectrum image on the right. The Combine Spectrum tool is shown on the far right.