The short answer is: use low and high to define which Canny edge responses enter the blur analysis, use radius to define how far around an edge the filter searches for local maxima, and use block_width, block_height, and block_pct to limit where the analysis is performed. Analyze the whole frame when the frame is the decision unit; use blocks when a local region matters. Keep all of these settings identical before comparing readings from different clips.
Practical decision and steps
The official blurdetect documentation describes low and high as Canny thresholds. high identifies strong edge pixels; low identifies weaker pixels that can be connected to them through 8-connectivity. Both values must be between 0 and 1, with low no greater than high. The documented defaults are 20/255 and 50/255.
radius controls the search distance around an edge pixel for local maxima. It changes the neighborhood considered by the blur calculation, not the image dimensions. Block options change the analysis area: positive block_width and block_height divide the image into blocks, while either dimension below 1 disables blocks and processes the whole image. block_pct restricts the calculation to the most significant blocks by percentage; the documentation gives block_pct=80:block_width=32:block_height=32 as an example.
Use this sequence:
- Define the question. Choose full-frame analysis for an overall frame reading, or block analysis when a face, title, product, or other local area is the actual concern.
- Record
low,high,radius,block_width,block_height,block_pct, andplanes. Do not infer a missing value from a colored node or interface label; use the actual filter settings and media-field labels. - Establish one baseline with the documented defaults and no blocks. Change one setting group at a time when investigating a difference.
- For a local diagnostic, choose a block size that matches the detail scale you want to inspect, then record which percentage of significant blocks is included. Treat the result as a different measurement scope, not as a better version of the full-frame result.
Acceptance worksheet
The following is an author-proposed comparison method, not a software test or a guarantee of expected values. Imagine the same 1920×1080 frame contains a sharp title in the center and a soft background. Record the observed output from each run rather than assuming the block setting will improve it.
| Run | Assumed settings | Observation to record | Decision |
|---|---|---|---|
| A | Defaults, no blocks | Full-frame reading includes title and background | Baseline for this frame and configuration |
| B | block_width=32, block_height=32, block_pct=80 | Reading reflects only the selected significant blocks under the filter’s rule | Use only for local or salient-content comparison |
| C | Same as B, but a different radius | Note whether the reading changes with the search neighborhood | Do not compare C with A as if only content changed |
A useful worksheet also labels the clip, frame or interval, pixel format, selected planes, and every filter option. If two readings differ, first ask whether the analysis scope or configuration changed; only then investigate a content difference.
Limits and next step
blurdetect is a no-reference blur metric that does not alter the input frames. Its reading is not a subjective quality score, proof that detail is readable, or delivery-acceptance result. A high or low value alone does not establish whether a viewer prefers the image, whether text meets a creative brief, or whether an upload pipeline will preserve it. The official description explains the filter options, but it does not define a universal pass threshold, a complete block-selection algorithm, or a rule that one radius is correct for every image.
If a result is unexpected, rerun the same frame with the baseline configuration, verify the actual filter string and selected planes, then inspect the relevant region visually at the intended viewing size. If the decision concerns delivery, review the encoded output separately. Do not claim that changing a threshold restores detail that is absent from the source.
Three quick questions
Should I always use blocks? No. Use them when the decision concerns a region or salient content; use whole-frame processing when the frame itself is the unit of judgment.
Does a larger `radius` mean more blur is detected? Not automatically. It changes the search neighborhood, so the reading may change, but the documentation does not promise a monotonic relationship.
Can I compare readings from two videos? Only when the relevant frame context and analysis settings are matched. Otherwise, label the difference as a configuration or scope difference before treating it as a content difference.
Sources
Sources and next step
This guide interprets public documentation and uses explicitly hypothetical acceptance examples, not application tests. Controls and availability can change; verify the cited manual against your installed version.
For work created in Imagild, carry the input requirements and acceptance checklist into the film workspace. The external software controls discussed above belong to their respective products.