SIPS Compliance Matrix

Tracking implementation status of algorithms defined in the NGA Softcopy Image Processing Standard (NGA.STND.0014 v2.4, 2019-08-21).

This is a living document. It is updated as operators are implemented and validated. The goal is incremental progress toward full compliance, with clear documentation of what is and isn’t covered at each release.

Reference

  • Standard: NGA.STND.0014_2.4 — Softcopy Image Processing Standard

  • Version: 2.4 (2019-08-21)

  • Source: osml-imagery-io/reference-materials/SIPS/SIPS_v24_21Aug2019.pdf

Status Legend

Symbol

Meaning

✅

Implemented and validated against SIPS verification matrices

🔨

In progress

📋

Planned for current release

📅

Planned for future release

➖

Not planned (out of scope or deferred indefinitely)

Algorithm Compliance Matrix

#

SIPS Algorithm

Section

Status

Toolkit Module

Notes

1

RRDS Generation

2.2

📋 v2.0

sips_resample.sips_rrds_resample()

7×7 anti-alias kernel + 4×4 LaGrange interpolation + 2× subsample. Validates against Tables 2.4–2.6.

2

Four-Point Interpolation

2.3

📋 v2.0

sips_resample (internal)

LaGrange 4×4 kernel with 1/32 pixel quantization. Compromise coefficients (Table 2.8). Used by RRDS and Scaling. Validates against Tables 2.9–2.15.

3

Convolution and Correlation

2.4

📋 v2.0

convolution.sips_convolve(), convolution.sips_correlate()

Mirror Edge - Odd boundary handling via cv2.BORDER_REFLECT_101. Validates against Tables 2.25–2.26.

4

1-D Look-Up Table

2.5

📋 v2.0

lut.apply_lut()

Per-band LUT application. Supports 8, 11, 12, 16-bit. Validates against Table 2.29.

5

Dynamic Range Adjustment

2.6

📋 v2.0

dynamic_range_adjustment.dynamic_range_adjust()

Histogram percentile-based clipping (Equations 2.16–2.20). DRAParameters.from_counts() implements the A/B modifier logic. Validates against Tables 2.32–2.34.

6

Scaling

2.7

📅 v2.1

sips_resample.sips_scale_resample()

Intra-octave scaling for non-power-of-2 factors. Uses 1-D anti-aliasing kernels derived from RRDS filter LSF + LaGrange interpolation. Validates against Table 2.43.

7

Asymmetric Pixel Correction

2.8

📅 Future

—

Resamples non-square pixels to square pixel space via spatially-variant LaGrange correlation. Uses the same interpolation engine as RRDS. Niche use case.

8

Color Management Module

2.9

📋 v2.0 (partial)

color_space.color_space_transform()

v2.0 delta: Hardcoded transforms for well-known color spaces (sRGB, ProPhoto RGB, AdobeRGB, CIE XYZ, CIELab). No ICC profile file parsing. Validates against Table 2.44 for sRGB→ProPhoto RGB. Full ICC support: post-v2.0 optional extra.

9

Max-Pixel Decimation

2.10

📅 v2.1

—

Per-channel max from 2×2 blocks for colorized change composites. ~20 lines. Low priority — specialized for change detection products.

10

Band Dependent Linearization

2.11

📅 v2.1

—

Per-band black/white point normalization for multiband imagery. Related to DRA but multiband-aware. ~50 lines.

11

LUT from Piecewise Math Formulae

2.12

➖

—

Utility for generating TTC tables from piecewise functions. Not needed for image display.

v2.0 Compliance Summary

What v2.0 Covers

  • RRDS Generation (the core pyramid quality algorithm) — full compliance with SIPS Section 2.2, validated against reference matrices

  • Four-Point Interpolation — full compliance, used internally by RRDS

  • Convolution and Correlation — full compliance with boundary handling

  • 1-D LUT Application — full compliance

  • Dynamic Range Adjustment — full compliance with auto-DRA algorithm

  • Color Management — partial compliance (well-known color spaces only)

Known v2.0 Deltas from Full SIPS Compliance

  1. CMM does not parse ICC profile files. Transforms are limited to hardcoded well-known color spaces. Users with custom sensor ICC profiles cannot use them through our CMM. Workaround: use Pillow’s ImageCms or pylcms2 externally, then pass the transformed array to our functions.

  2. Intra-octave scaling not implemented. Only power-of-2 downsampling is supported via sips_rrds_resample(). Smooth zoom between pyramid levels uses OpenCV resamplers as an interim. Planned for v2.1.

  3. No asymmetric pixel correction. Images with non-square pixels are not automatically corrected. The LaGrange interpolation engine needed for this is implemented (shared with RRDS), so adding this is straightforward when needed.

  4. No max-pixel decimation. Colorized change composite products use standard RRDS instead of the max-pixel algorithm. Planned for v2.1.

  5. No band dependent linearization. Multiband imagery uses per-band DRA instead of the BDL algorithm. Planned for v2.1.

  6. SIPS databases not bundled. The SIPS spec is accompanied by databases of kernels (MTFR, MTFE, anti-aliasing, TTC families). We implement the reference kernels from the spec text but do not bundle the full SIPS database files. Users with access to the SIPS databases can load them via apply_lut() and sips_convolve().

Verification Approach

Each implemented algorithm is validated against the SIPS verification matrices provided in the specification:

Algorithm

SIPS Verification Data

Our Test

RRDS Generation

Tables 2.4 (R₀), 2.5 (R₁), 2.6 (R₂)

Exact matrix comparison

Four-Point Interpolation

Tables 2.10–2.15

Single-point and 10×10 matrix verification

Convolution

Tables 2.17–2.26

5×5 and 4×4 kernel examples, both odd and even

1-D LUT

Tables 2.27–2.29

TTC Family 0 Member 0 verification

DRA

Tables 2.31–2.34

10×10 11-bit matrix with known parameters

Scaling

Tables 2.35–2.43

0.4× absolute magnification example

CMM

Table 2.44

sRGB→ProPhoto RGB conversion

Error tolerances follow SIPS Section 4, Table 4.1:

Bit Depth

Max Absolute Error

8-bit

±1 digital count

11-bit

±1 digital count

12-bit

±1 digital count

16-bit

±1 digital count

Reference Image Chain Coverage

The SIPS reference image chain (Section 3) defines the processing order for single-band and multiband imagery. Here is our coverage of each element:

Single-Band Reference Chain (Figure 3.8)

Sharpness Enhancement → DRA → 1-D LUT (TTC)
        ↓                 ↓         ↓
   MTFC/MTFE d/b     Histogram   TTC d/b

Element

v2.0 Status

Notes

Sharpness Enhancement

✅ via sips_convolve()

User supplies MTFC/MTFE kernel

Histogram

✅ via compute_statistics()

DRA

✅ via dynamic_range_adjust()

1-D LUT (TTC)

✅ via apply_lut()

Multiband Reference Chain (Figure 3.9)

DRA → CST → [ACS Operations] → CST → Output Preparation
 ↓     ↓                              ↓
DRA  ICC Profile                    ICC Profile
d/b    d/b                            d/b

Element

v2.0 Status

Notes

DRA

✅ via dynamic_range_adjust()

Per-band

CST (Color Space Transform)

📋 partial

Well-known spaces only

ACS Operations (sharpening, LUT)

✅ via sips_convolve(), apply_lut()

Output Preparation

📋 partial

Quantization to output bit-depth; CMM limited to well-known spaces

Non-Interactive Processing (Figure 3.3)

Element

v2.0 Status

Notes

Band Equalization

➖

Sensor-specific; not in scope

MTF Restoration

✅ via sips_convolve()

User supplies MTFR kernel

Atmospheric Correction

➖

Sensor-specific; not in scope

Data Remapping (PEDF/LinLog)

📅 v2.1

Equations 3.1–3.4 in spec

Asymmetric Pixel Correction

📅 Future

RRDS Generation

📋 v2.0