Five-Tier Input Domain Resolution & Unified Density Pipeline

A recurring design flaw across legacy film scanning and negative inversion software is allowing the input file container to dictate downstream optical density mathematics. Under such fragmented architectures, digital camera RAW captures undergo heuristic exposure weighting, Hasselblad Flextight or Imacon FFF files follow proprietary cropping logic, and standard TIFF files are routed through uncalibrated display gamma curves. This tight coupling between file formats and inversion algorithms introduces inconsistent color renditions across different digitizers, undermining the reproducibility of photographic color science.

In version 1.1.0, the NexFilm Engine establishes an architectural principle: the input file format determines only the input domain, while the underlying density mathematics remains globally unified.

Whether the incoming source is a camera lossless RAW file, a high-bit-depth LinearRaw scanner DNG, an FFF container embedding proprietary metadata, or a 16-bit TIFF with an embedded color management profile, the decoder responsibility is strictly confined to identifying chromatic primaries and electro-optical transfer characteristics. The decoded data is projected losslessly into a standardized 32-bit floating-point Linear ProPhoto workspace. Once linearized, all subsequent optical density operations—including film base subtraction, printer lights trimming, bounded channel span response, and display mapping—execute through identical mathematical pathways.

Input File (RAW / FFF / TIFF / DNG / JPEG)
       │
       ▼
resolve_input_domain(path, scanner_profile)
       │  ← The sole permissible branching point
       ▼
Unified Workspace: Linear ProPhoto f32 (eliminating u16 quantization loss)
       │
       ▼
Unified Density Transform: D = -log10(max(T, epsilon))
       │
       ▼
Channel-Wise Base Subtraction: D_net = D_raw - D_base (shared origin, zero offset)
       │
       ▼
Bounded Channel Span Response (1.15x deadzone / 1.35x cap / 3.0x truncated cap)
       │
       ▼
Positive Linearization & Display Mapping: ProPhoto -> Output Gamut & Target LUT

This clean decoupling eliminates the gamut clipping and precision loss caused by legacy intermediate quantizations into 16-bit unsigned sRGB, ensuring that wide-latitude film negatives maintain radiometric consistency across all capture hardware.

To translate heterogeneous image captures into physical relative transmittance, NexFilm implements a prioritized resolution chain encapsulated in the resolve_input_domain function. The engine evaluates five tiers of metadata evidence in sequence, attaching provenance records, color primaries, transfer functions, and confidence flags to each processed frame:

1. Tier 1: Embedded ICC Color Profile (Verified)
   If a TIFF or JPEG image embeds a valid ICC profile (such as Adobe RGB (1998), Display P3, or custom scanner calibration profiles), the engine extracts chromaticity matrices (rXYZ, gXYZ, bXYZ) and parametric tone reproduction curves. This profile is designated as Verified and carries the highest authority.
2. Tier 2: Scanner Container Metadata & Property Lists (Verified / Estimated)
   For Hasselblad Flextight / Imacon .fff files, the decoder inspects the TIFF header for embedded XML property lists (plist), extracting RGBProfile descriptors and Gamma coefficients. If the container explicitly designates AdobeRGB or scanner-calibrated profiles, the declared gamma expansion applies directly. When container properties are absent, the engine falls back to documented scanner standards (Gamma 1.8, sRGB primaries, tagged as estimated).
3. Tier 3: RAW / LinearRaw Metadata & Sensor Matrices (Declared)
   For camera RAW files (such as CR2, NEF, ARW), the engine parses camera native color filter arrays and white balance multipliers through LibRaw. For DNG scans produced by specialized film scanners, the parser evaluates SubIFD tags; if PhotometricInterpretation = 34892 (LinearRaw) is identified without lossy compression, the samples are treated as unscaled linear transmittance.
4. Tier 4: Documented Scanner Model Defaults (Documented Default)
   When a user explicitly selects a hardware scanner profile (such as Nikon Super Coolscan 9000 ED or Minolta Dimage Scan Elite 5400) and the file lacks embedded ICC tags, the engine applies documented spectral primaries and transmission characteristics calibrated for that model.
5. Tier 5: sRGB Fallback with Diagnostic Warning (Estimated Fallback)
   If all preceding criteria fail to match, the image data is interpreted under standard sRGB assumptions. The processing pipeline marks the record with an estimated diagnostic flag, alerting the user to review color accuracy in the inspection panel.

Resolution outcomes are permanently recorded in the project SQLite database under PipelineProcessingReport::input_domain, detailing primary coordinates, transfer functions, and diagnostic logs.

The measurement accuracy of the unexposed orange film base establishes the white balance anchor for the entire negative inversion. Heuristic software often locates the brightest overall pixel in the image and designates it as the base reference. When applied to scans containing perforations, backlight direct glare, or expansive light-colored backgrounds, this simplistic assumption introduces significant color contamination and flat contrast.

NexFilm implements a three-tier base sampling hierarchy governed by strict physical validation thresholds:

During candidate validation, the film_base_band_agrees routine enforces a unidirectional physical inequality: because the unexposed film base represents the maximum possible optical transmittance of the film strip, image-forming areas within the frame cannot physically transmit more light than the base itself. If the measured border narrow band is darker than the emulsion tail by more than 0.05 D, the engine concludes that the border is contaminated by flare or lighting falloff and automatically falls back to the internal low-density tail.

After deducting the measured base density, the net optical density $D_{\text{net}} = D_{\text{raw}} - D_{\text{base}}$ aligns to an analytical origin where the base density in red, green, and blue evaluates identically to zero.

However, in complex digitization workflows—particularly panoramic stitched TIFF files generated from multiple RAW captures (such as Lightroom panoramas _DSC7569-Pano.tif and _DSC7583-Pano.tif)—upstream merging operations often compress the red channel density span down to one-third or one-fourth of the green channel span. If a shared display window width is applied uniformly across all three channels, the compressed red channel is forced into the lower luminance register, causing cyan-green casts or blocked highlights.

While earlier experiments introduced per-channel display curves, that approach distorted the color balance of healthy negatives. NexFilm introduces the Bounded Channel Span Response model, establishing three mathematical safeguards:

1. Shared Window Origin: Whenever a valid film base is present, channel_offsets remains fixed at zero across all three channels. Consequently, regardless of span scaling, density zero (the clear film base) always maps to identical positive display white, preventing baseline color contamination.
2. 1.15x Deadzone Threshold: For standard negatives processed and scanned under balanced lighting (where channel span ratios remain between 1.00 and 1.15), channel gain adjustments remain in the deadzone, preserving identical three-channel window allocation.
3. Bounded Gain Ceilings: Using measure_content_channel_spans, the system measures effective density spans between the 2nd and 98th percentiles per channel. When imbalance exceeds 1.15, window widths are scaled via a smooth linear ramp: $$g_i = \text{clamp}\left(\frac{\text{span}_i}{\text{luma}(\text{span})}, \frac{1}{g_{\text{max}}}, g_{\text{max}}\right)$$ For conventional scans, the gain ceiling is capped at $g_{\text{max}} = 1.35$. Only when severe channel collapse exceeds a 2.0 span ratio does the engine identify an upstream truncated channel, moderately relaxing the ceiling to 3.0.

This formulation rejects the arbitrary color leveling of the traditional gray-world hypothesis (which forces average scene RGB to neutral gray). It preserves the natural atmospheric palette of the scene while recovering dynamic range for damaged channels.

The stability of the unified density pipeline is verified through automated regression suites. The table below compiles pixel-level measurements executed via density_pipeline_sample_report across standardized real-world film captures, comparing whole-image red-to-green and blue-to-green channel ratios between legacy v1.0.2 processing and the unified pipeline:

These benchmark measurements confirm that for healthy captures, the unified pipeline matches the theoretical physical baseline with zero pixel divergence. For severely compressed panoramic composites, the bounded span response restores usable dynamic range without compromising the neutral film base.