Sharpening with BlurXTerminator and Noise Reduction with NoiseXTerminator

BlurXTerminator performs astronomical deconvolution: reconstruction intended to reduce blur. NoiseXTerminator estimates and reduces noise. A sharper appearance and a smoother appearance are different goals, and neither alone proves that the image has improved.

Both tools are developed by RC Astro and require the appropriate purchased or trial licenses. Use the installation guide under Continue learning if they are not ready. Record the process and model versions: the linked NoiseXTerminator manual specifically describes the 2/AI3 controls, and another release may differ.

This exercise uses a linear broadband RGB image with stars still present. It aims for a restrained processing checkpoint, not a final sharpening or smoothing prescription for every target.

Save the calibrated, integrated source before beginning. Complete the relevant background and color-calibration work for your chosen workflow. STF may be active for viewing; it does not make the pixels nonlinear.

Use BlurXTerminator on linear data before noise reduction and, for this introductory workflow, before star removal. NoiseXTerminator can work on linear or stretched images; here it follows the accepted sharpening result while the image is still linear.

If planning a separate Correct Only pass, consult the vendor’s recommendation to place that correction before precision color calibration such as SPCC. Do not insert another correction pass automatically into an image that already received one. Keep correction, calibration, and sharpening checkpoints identifiable.

Make a working duplicate named Target_RGB_BXT_trial01. Keep the original open for comparison and check that no unintended mask is attached.

Correct Only focuses on correcting supported aberrations rather than applying the normal sharpening settings. It is not the same as sharpening with a low amount, and it is not required as an extra pass every time.

For a normal sharpening trial, leave Correct Only off. Sharpen Stars and Sharpen Nonstellar control different features; Adjust Star Halos changes stellar surroundings. Start with restrained amounts and neutral halo adjustment rather than maximizing every control.

Automatic PSF estimates the point spread function—the image’s blur profile—from available information. Stars provide important references. Sparse-star or starless regions need extra care; manual PSF use should be based on a measured stellar FWHM, not an invented diameter chosen because it looks sharper.

Use the installed manual for exact control names and ranges in your version. The purpose of this first trial is to find a useful, modest change.

Pick three representative regions: a bright star with its surroundings, a high-signal structure, and a faint area with real texture. Revisit the same locations for every trial rather than comparing whichever feature happens to look best.

Look for dark rings, brittle edges, repeated worm-like patterns, or fine structures that suddenly look equally sharp in bright and very noisy regions. A striking result at fit-to-window can still have distracting artifacts at 1:1.

Compare with the source before deciding that a newly crisp knot is genuine detail. Lower the relevant sharpening amount if the improvement is uncertain. Do not smooth a questionable sharpening artifact away and then count it as a successful reconstruction.

A more modest result is worth keeping when it preserves natural transitions and gives a clear improvement over the source. You can also keep the original if the data do not support a useful gain.

Make Target_RGB_NXT_trial01 from the accepted BXT checkpoint. This keeps the sharpening comparison separate from the noise-reduction decision.

In the 2/AI3 model, Iterations controls internal successive-approximation steps; it does not mean you should repeatedly apply the process to its own output. More steps can help some noisy areas but cost time and can introduce artifacts.

Intensity/color separation allows different treatment of brightness noise and color noise in a color image. Frequency separation distinguishes finer noise from broader variations. Use either only after the simple-mode comparison gives you a reason to explore it.

Broad fluctuations may be real dust or unresolved structure, not noise. Compare with the source when testing low-frequency reduction. Return to simple mode if you cannot identify a consistent benefit.

Older screenshots may show a Detail control. In the linked manual, that belongs to AI2 and does not operate with AI3. Match instructions to the selected model rather than trying to reproduce every control from an old tutorial.

Keep three views available: the pre-BXT source, accepted BXT result, and accepted NXT result. Use identical display settings when comparing linear checkpoints so that an independently recomputed STF does not disguise what changed.

Compare source to BXT for sharpening, BXT to NXT for noise reduction, and source to NXT for the overall effect. Check that the smoother version still contains the faint structures you wanted to protect.

Finish by inspecting the whole image and the same 1:1 regions. Look for uneven texture, halos, color blotches, and transitions between bright and faint areas. Save the process instances and record the settings and model versions that produced the retained checkpoints.

Already stretched or noise-reduced input for BXT: return to an earlier linear, non-denoised checkpoint. Renaming a file or turning STF off does not reverse permanent processing.

Over-sharpened faint structure: lower the sharpening, review whether stars adequately support Automatic PSF, and consult the manual before attempting a measured manual PSF. Do not increase a guessed diameter to force more detail.

Noise reduction removes faint texture: compare a lower Denoise setting from the same BXT source. Review low-frequency settings if enabled. A uniform-looking background is not automatically a more accurate image.

New rings or patches: establish whether they appeared after BXT, NXT, or an earlier operation by comparing the checkpoints. Change one stage at a time rather than adding a compensating pass to the finished trial.

Slow processing or a download error: check model readiness and the publisher’s current hardware guidance. Keep exact error text when asking for support; do not share license keys.

Keep the accepted linear checkpoints and the untouched integration. You can now follow the star-separation lesson, or continue to a permanent stretch when appropriate for the workflow. If you denoise a starless branch instead, retain the pre-denoise starless source for a similarly controlled comparison.

The example values here define an original beginner exercise, not tested settings for every camera or target. Use the current installed documentation when labels or models differ from the linked manuals. The goal is a defensible improvement you can see and explain, with enough source information saved to revise it later.