Target Morphology Strategy

The core principle

Object type should change what NOVA protects, measures, and tests. It should not turn into an unquestioned preset recipe. A galaxy can have a bright nucleus and faint outer arms; processing that wins by darkening the sky can silently erase that outer structure. An emission nebula's visible signal can occupy most of the frame, so a background or contrast method can mistake target signal for sky. A reflection nebula's color is scattered starlight, not emitted light, so treating its blue dust as a color cast to neutralize is a category error. A planetary nebula can be tiny in the frame with fine shells and a bright core, so whole-frame metrics can be nearly blind to real damage inside it. A globular cluster is dominated by a dense, crowded core, so star shape and halo preservation are the central risk. An open cluster is more loosely crowded and can sit inside surrounding nebulosity that a single-label classification misses entirely. None of this requires a wholly different pipeline per class -- it requires knowing what failure would be most expensive for this kind of object.

The seven current object-type buckets

Ontology v1.3 defines seven public object-type buckets, each with a short list of priority qualities and a stretch preference: galaxy (color balance, detail level, dynamic range; GHS), emission nebula (noise, gradient, detail level; Statistical Stretch), reflection nebula (color balance, gradient; Veralux), globular cluster (star roundness, noise; Statistical Stretch), planetary nebula (detail level, star roundness; GHS), open cluster (star roundness, color balance; Statistical Stretch), and unknown (overall; Statistical Stretch). This is a useful small intent vocabulary, but it badly understates what object type actually changes at runtime -- see the next section -- and its named stretch preferences are current NOVA priors, not physically established rules; the Stretch article's own evidence-strength labeling remains the authority on each engine's actual comparative evidence.

What object type actually touches

The ontology's compact per-type table is not the runtime strategy. Beyond the three listed priorities, object type currently also changes MLT's working scale for galaxies and nebulae relative to other classes, BlurXTerminator's nonstellar sharpening strength (higher for galaxies, lower or default elsewhere), HistogramTransformation's target-background priors, Statistical Stretch's target range (darker for galaxies), STF's target background (lower for galaxies and clusters), GHS's alpha (higher for galaxies), color saturation's computed boost (higher for nebula classes), iHDR/HDRMT aggressiveness (more for nebulae, less for clusters), LHE/CLAHE local-contrast aggressiveness (lower for smooth nebular structure, more permissive for galaxies), and dark-structure enhancement (boosted for galaxies, reduced for clusters). Treating morphology strategy as only 'priority qualities plus a stretch engine' misses most of what it actually does.

Type is often a guess, not a measurement

When a run has no explicit object type, NOVA infers one from the target's name or catalog identifier. That is a practical routing mechanism, not an astronomical classifier, and it can be wrong for mixed fields (a cluster with surrounding emission or reflection nebulosity), galaxy groups or interacting systems, generic names with no recognizable class token, targets whose catalog identity does not match the visually dominant structure in the crop, or mosaics where the named target occupies only part of the field. A wrong inferred type does not necessarily make a run invalid, but every morphology-derived setting and every historical prior consulted for that run was applied under the wrong hypothesis -- which is why the *effective* object type used, not merely the target name, is what provenance needs to record.

Historical priors are partitioned, not proven

NOVA's Experiment Mode looks up historical priors by both processing step and object type, so a galaxy result is not pooled blindly with a globular-cluster result. That is a directionally sound partition, but it does not repair the underlying evidence problems this Handbook already names elsewhere: historical win fraction is still not statistical confidence, morphology partitioning does nothing to correct for selection bias, repeated non-independent runs on the same target, or drift in code, candidate set, or evaluator between runs.

The globular cluster case study

This is not a hypothetical risk. Preserved run history for globular clusters records real disagreement rather than a settled preset: an earlier globular curve strategy had favorable M13/C80 experience behind it, but a later M12 run showed that same style of curve darkening the halo while brightening upper percentiles, and an M3 run showed a globular-core-rolloff curve crushing the sky while leaving the core near the top of the range. The correct conclusion from this evidence is not 'globulars prefer curve X' -- it is that globular morphology creates predictable risks (crowding, core compression, star-shape damage, halo loss) that justify cluster-specific experiments and protected regions of interest, evaluated fresh each time rather than assumed settled.

What each morphology makes expensive to damage

Galaxy -- core-to-outer-structure dynamic range, fine detail in arms and dust lanes where SNR supports it, faint structure near the sky floor, and broadband color balance; watch for a darker background quietly erasing the outer halo or arms.

Emission nebula -- low-surface-brightness diffuse and filamentary structure that can occupy most of the frame; watch for background or gradient methods mistaking real field-filling nebulosity for sky, and for noise reduction erasing genuine faint filaments alongside noise.

Reflection nebula -- faint dust and smooth low-surface-brightness structure whose broadband color is scattered starlight, not emitted light; watch for gradient extraction removing dust that fills the frame, and for blue/cool color being misread as a cast to neutralize.

Planetary nebula -- small-scale nonstellar structure, central-star and shell integrity, and local dynamic range in a target that can be a tiny fraction of the frame; watch for whole-frame metrics being nearly blind to real damage inside the target itself.

Globular cluster -- star roundness, crowded-core separation without ringing or false deblending, controlled core brightness, and halo preservation; watch for a darker sky improving apparent contrast while deleting real halo stars, and for raw star count or lower FWHM rewarding fragmentation or ringing rather than genuine resolution.

Open cluster -- star shape and color, a clean but not artificially crushed background, and less aggressive core compression than a globular cluster deserves; watch for surrounding reflection or emission nebulosity (the Pleiades is the standing example) making a pure cluster treatment incomplete.

Unknown/other -- a safe routing fallback, not a morphology claim; evidence produced under 'unknown' should stay visibly tagged as such, since its adaptive parameter path can differ from a correctly classified target.

Confounders that break a morphology claim

A claim like 'galaxies benefit from stronger GHS' or 'globulars prefer Statistical Stretch' needs more than a pile of target wins to be credible. The recurring traps are: capture/filter confounding, since emission-nebula targets skew heavily toward LP or narrowband capture, so an apparent morphology preference can actually be a filter/channel-mapping preference; target-size/occupancy confounding, since a tiny planetary nebula and a frame-filling emission nebula give whole-frame metrics radically different sensitivity even though both are labeled 'nebula'; SNR/integration confounding, since a morphology class can correlate with typical integration time, so a denoise or sharpen win may reflect SNR rather than morphology; scale/resampling confounding, since native, drizzled, and mosaicked images give the same pixel-valued radius or tile parameter a different angular meaning; adaptive-tuning confounding, since NOVA computes parameters from both measurements and object type together, so a nominal method comparison can secretly be a tuning-policy comparison; name-classification confounding, since inferred-type error can contaminate a learned morphology prior at its source; and repeated-target dependence, since ten runs of the same galaxy are not ten independent replications of 'galaxy' as a class.

Mixed fields break a single label

The Pleiades is the clean public example of why a single categorical type cannot fully describe processing risk: it is simultaneously an open cluster and a reflection-nebula field, and a treatment aimed purely at star contrast will lose the surrounding dust. The same problem recurs for a galaxy with significant integrated flux nebulosity, a globular or open cluster with a background emission complex behind it, or a compact nebula embedded in a dense stellar field. Where a mixed field is recognized, the right response is to protect both the star and diffuse-signal regions of interest rather than commit to one class's recipe.

Stronger and weaker uses of morphology

Morphology is on solid ground when it changes which regions or features get protected, which failure modes count as expensive (ringing, star merging, faint-signal loss, gradient overfit, core clipping), which metrics and regions of interest carry weight, which candidate families are worth testing at all, and what a reasonable starting range looks like before measured image state constrains it further. It is on much weaker ground whenever it is used alone to justify one universally best stretch engine, one fixed target sky level, one sharpening or denoise strength, one saturation amount, one HDR iteration count, one CLAHE setting, one curves preset, one background-extraction model, or one palette/color treatment -- those decisions depend on measured SNR, PSF, scale, target occupancy, capture/filter type, linear/nonlinear state, clipping, background structure, engine/version, and prior processing, not on the object-type label alone.

How this page is meant to be used

This page does not duplicate tool instructions -- it routes to them. Stretch, Background Extraction, Color Calibration, Deconvolution, Denoise, and Star Correction each define their own methods, evidence, and failure modes; morphology only changes which of their concerns matter most for a given target and which regions of interest should be checked before accepting a result. If a claim here and a process article's own evidence ever disagree about a specific tool's comparative performance, the process article's evidence-strength labeling governs, since it is the article actually scoped to evaluate that tool.

Auditable claims

1. Ontology v1.3's seven object-type buckets each list only two or three named 'priority qualities' and one stretch preference, but current NOVA code applies object type far more broadly than that summary implies -- among other operations, it changes MLT scale, BlurXTerminator nonstellar sharpening strength, HistogramTransformation target-background priors, Statistical Stretch's target range, STF's target background, GHS alpha, saturation boost, iHDR/HDRMT aggressiveness, LHE/CLAHE aggressiveness, and dark-structure-enhancement strength. The three-item ontology table is a small intent vocabulary, not a complete description of what object type actually does at runtime.

   source-confirmed · none

2. When no explicit object type is supplied, NOVA infers one from the target's name or catalog identifier -- a practical routing heuristic, not an astronomical classifier. It can misclassify mixed fields (a cluster embedded in nebulosity), galaxy groups or interacting systems, generic or unusual names, and mosaics where the named target occupies only part of the frame. A wrong inferred type does not necessarily invalidate a run, but it means every morphology-derived setting and prior was applied under the wrong hypothesis, so the effective type used -- not just the target name -- is what later analysis and provenance must record.

   source-confirmed · none

3. NOVA's historical Experiment Mode priors are queried by both step and object type, so a galaxy result is not blindly pooled with a globular-cluster result. That partitioning is directionally sound, but it does not by itself correct for selection bias, repeated non-independent targets, code or candidate-set version drift, or evaluator drift -- a morphology-partitioned historical win fraction is still not statistical confidence.

   source-confirmed · none

4. Preserved run history shows real disagreement rather than a settled globular-cluster preset: earlier M13/C80 runs recorded favorable experience with a particular curve strategy, but a later M12 run showed that same style of curve darkening the halo while brightening upper percentiles, and an M3 run showed a globular-core-rolloff curve crushing the sky while leaving the core near the top of the range. This is retained specifically because it is the concrete case against treating any morphology class as having one validated best method -- 'globulars prefer curve X' is not supported by this evidence; 'globular morphology creates crowding, core-compression, and halo-loss risks worth testing for' is what the evidence actually supports.

   source-confirmed · none