How to Audit Image Quality in Agisoft Metashape Before Alignment

More photographs do not always mean a stronger Agisoft Metashape project. A few blurred frames can weaken image matching, introduce soft areas into a texture and waste processing time. The useful question is not simply whether a photograph has a low score: it is whether that photograph provides coverage that no sharper image can replace.

What Metashape’s image quality value measures

Metashape estimates a quality value from the sharpness of the most focused portion of each image. In the Metashape Professional user manual, Agisoft recommends disabling images with a value below 0.5 before alignment and texture generation. Treat 0.5 as a screening threshold, not a universal definition of a usable photograph: a sharp foreground can produce a respectable score even when the asset of interest is out of focus.

A score also cannot tell you whether a frame is essential to bridge two flight lines, document an otherwise hidden facade or cover a small feature. Inspect the original image at 100% magnification and compare it with its neighbors before excluding it.

Run a repeatable photo audit before alignment

1. Keep the original files unchanged. Make a working copy of the project and record the image count and camera groups.
2. In the Photos pane, switch to Details view, select the relevant photographs, right-click and choose Estimate Image Quality. Inspect the resulting Quality column.
3. Sort by quality and review the lowest-scoring frames in context. Look for subject motion, camera shake, missed focus, severe noise, glare and exposure clipping.
4. Check each candidate against adjacent images and the planned coverage. Disable a bad frame only when adequate neighboring views remain.
5. Align the retained set. Inspect camera coverage, components and the locations of any unaligned photographs. Restore an excluded frame only if it contributes usable information.

Agisoft’s image capture guidance advises keeping source images unmodified for photogrammetry. Do not sharpen, resize or geometrically transform originals merely to raise a quality score; those edits can compromise calibration or conceal the cause of a defect.

Decide by defect and by role

Observation	Practical decision
Soft frame with many equivalent sharp neighbors	Disable it and document why.
Low score but the only view of a critical recess	Inspect at full resolution; keep as a provisional bridge if features are still measurable, and flag the area as uncertain.
Sharp ground but blurred elevated structure	Judge the structure itself; the global score may be misleading.
Good alignment, visibly soft texture	Review which source images contribute to texturing; an aligned frame is not necessarily a good texture source.
Blurred object moving through an otherwise sharp scene	Assess whether the object matters; local masking or a different source image may be appropriate.

Check whether the revised set actually improved the project

Keep a baseline copy and compare the same area using the same alignment settings. Record the number of aligned cameras, connected components, image coverage and independent check-point results if available. A lower average reprojection error alone does not establish better map accuracy. Our guide to the Metashape processing report explains why independent control matters, while the alignment troubleshooting guide helps diagnose a newly disconnected block.

If removing a weak image creates a gap, the strongest remedy may be another field visit. No software setting can invent a missing viewpoint. During the next acquisition, use a faster shutter where necessary, confirm focus on the actual subject, inspect a sample at full resolution and allow enough overlap to survive the loss of a few frames. For aerial work, design that redundancy alongside a realistic ground sampling distance.

Three checks that a single quality number cannot perform

Is the relevant plane in focus?

A facade can be soft while foreground trees are crisp. Zoom into the facade, not the most visually striking area. For close-range objects, inspect the front, sides and recessed details separately; shallow depth of field can make different parts of the same frame behave very differently.

Does the photo connect otherwise separate views?

Some weak frames are transitional images between aerial and terrestrial coverage. Disabling one may split the alignment into components. In that case, compare the resulting network with and without it. If the only bridge is genuinely blurred, retaining it does not make the measured geometry trustworthy; plan a reshoot with better cross-view overlap.

Will the photo be used in the final appearance?

A frame can help alignment yet soften a texture or orthomosaic. Define whether your deliverable is a measured surface, an image product or both. Inspect the final output at native pixel resolution, especially along high-contrast edges. This prevents a superficially good alignment statistic from becoming the sole acceptance test.

Suggested audit record: note image ID, quality value, visible defect, whether the subject is affected, replacement coverage, keep/disable decision and result after alignment. This short record is particularly useful when several people review a large project or a client asks why some of the original photographs were omitted.

Use a small pilot when the decision is ambiguous

On a difficult dataset, duplicate a representative chunk covering the same object and keep all important settings constant. Align once with the questionable frames enabled and once with them disabled. Compare connectivity, camera placement and check-point behavior, not only processing time or total tie-point count. If the two results are equivalent and the images are visibly defective, the cleaner input set is easier to justify. If one image is the only bridge between groups, treat the gap in acquisition as the central finding. Keep both pilot results so the exclusion rule can be applied consistently to the complete project.

Frequently asked questions

Should every image below 0.5 be deleted?

No. Agisoft gives 0.5 as a recommendation for disabling poor images, but removal should follow inspection of the subject and the remaining overlap. Disabling is reversible and preferable to deleting source evidence.

Can a sharp image still be unsuitable?

Yes. Reflections, repeated patterns, severe overexposure or a sharp background behind a blurred subject may cause problems that a sharpness-based score cannot fully describe.

Does a high quality score guarantee survey accuracy?

No. Image quality supports matching and visual detail. Geometric accuracy additionally depends on network geometry, calibration, reference data and independent validation.

Technical basis: Agisoft Metashape Professional user manual, Image quality section and Agisoft image capture guidance.