Model the Degradation Before Restoring
An old scan is blurred and noisy in different ways.
Read the full lesson and open its transcript below. A narrated video will be added after production.
What you will learn
An old scan is blurred and noisy in different ways.
A practical question
An old manuscript scan is both blurred and grainy. If we treat those as one problem, we may sharpen the grain while the writing remains unreadable. Let's describe what happened to the image before choosing a restoration tool.
The useful idea
Blur spreads the contribution of a point into neighboring pixels. Camera motion can stretch that spread in one direction; defocus can spread it differently. Noise adds variation to the measurements. Sampling and compression may add further limits. Look at known sharp boundaries and quiet paper areas to estimate which effects are present.
Work through the example
Compare a trail beside a dark letter with the soft edge of a defocused letter. The trail's direction can suggest motion, while a more symmetric softness may suggest defocus. These are clues, not certainty. If we assume the wrong blur, an algorithm can produce crisp-looking strokes that were never on the page. Start with the shape of the damage in the raw frame. A one-direction trail after camera movement differs from the softer spread caused by defocus. Estimate the cause from edges and acquisition conditions, then test whether that model predicts several examples, not just one attractive restoration. If the assumed blur is wrong, an algorithm may invent a crisp-looking edge in the wrong place. Keep the original beside every restored candidate, and describe the uncertainty before using restored pixels as measurement evidence. Now check the idea on a concrete example: Estimate whether camera motion or defocus better explains an edge trail. A possible trap is worth naming: The wrong model can produce a plausible but misleading reconstruction. The answer follows from that evidence: A degradation model separates blur, noise, and sampling assumptions before choosing a restoration method.
Limits and next step
Keep the raw scan, document the proposed degradation model, and test it where we know the answer. Before we attempt to undo blur, we'll estimate how much of the background variation is noise.
Worked example
Estimate whether camera motion or defocus better explains an edge trail.
Quick check: test your understanding
Try it yourself: Consider this case: Estimate whether camera motion or defocus better explains an edge trail. Explain how you would check the result. In your answer, address this limitation: The wrong model can produce a plausible but misleading reconstruction.
Check the reasoning
A degradation model separates blur, noise, and sampling assumptions before choosing a restoration method. The wrong model can produce a plausible but misleading reconstruction.
What is the main idea behind model the degradation before restoring?
A degradation model separates blur, noise, and sampling assumptions before choosing a restoration method.
How is it used in the example?
Estimate whether camera motion or defocus better explains an edge trail.
What limitation should be checked?
The wrong model can produce a plausible but misleading reconstruction.
Further reading
Digital Image Processing, 4th edition — Rafael C. Gonzalez and Richard E. Woods.
Original ToolsInVision teaching examples. Editorial approach · Suggest a correction
Accessible referenceLesson 45 transcript311 words · Narration ready for the upcoming videoOpen transcript
An old manuscript scan is both blurred and grainy. If we treat those as one problem, we may sharpen the grain while the writing remains unreadable. Let's describe what happened to the image before choosing a restoration tool.
Blur spreads the contribution of a point into neighboring pixels. Camera motion can stretch that spread in one direction; defocus can spread it differently. Noise adds variation to the measurements. Sampling and compression may add further limits. Look at known sharp boundaries and quiet paper areas to estimate which effects are present.
Compare a trail beside a dark letter with the soft edge of a defocused letter. The trail's direction can suggest motion, while a more symmetric softness may suggest defocus. These are clues, not certainty. If we assume the wrong blur, an algorithm can produce crisp-looking strokes that were never on the page. Start with the shape of the damage in the raw frame. A one-direction trail after camera movement differs from the softer spread caused by defocus. Estimate the cause from edges and acquisition conditions, then test whether that model predicts several examples, not just one attractive restoration. If the assumed blur is wrong, an algorithm may invent a crisp-looking edge in the wrong place. Keep the original beside every restored candidate, and describe the uncertainty before using restored pixels as measurement evidence. Now check the idea on a concrete example: Estimate whether camera motion or defocus better explains an edge trail. A possible trap is worth naming: The wrong model can produce a plausible but misleading reconstruction. The answer follows from that evidence: A degradation model separates blur, noise, and sampling assumptions before choosing a restoration method.
Keep the raw scan, document the proposed degradation model, and test it where we know the answer. Before we attempt to undo blur, we'll estimate how much of the background variation is noise.