17-EFT.WP.Methods.Imaging v1.0 | Chapter 15 Use Cases and Reference Implementations

Chapter 15 Use Cases and Reference Implementations

One-Sentence Goal
Provide executable blueprints for three representative scenarios—offline batch, online real-time service, and event/ToF fusion—covering the full loop from acquisition → modeling → calibration → imaging → QC → release, with binding to P/S/M/I, SLI.img.*, and SLO.*.

I. Scope & Targets

1. Inputs
   • Raw or linearized data y_raw or y_lin, multi-modality sequences Y = { y_k }, and device/mode descriptors Device, Mode (Chapter 3).
   • Calibration & QC fixtures: flat/dark frames, color charts, geometric boards, slanted-edge targets, point/line sources.
   • Metadata & reference conditions: RefCond, unit(x), dim(x), ts, tau_mono, offset/skew/J, gamma(ell).
2. Outputs
   Published imaging artifacts D_img.clean, quality panel SLI.img.*, compliance report and signatures manifest.imaging.*, and audit_report.
3. Constraints
   • Evaluate metrics and perform calibration in the linear radiometric domain (Chapter 4).
   • Compute arrival time using both forms (Methods.Cleaning v1.0, Ch. 6).

II. Terms & Variables

• Domains & symbols: Omega (spatial), f (frequency), t (time), ell (path).
• Key quantities: MTF(f), MTF50, NPS(f), PRNU, DSNU, DeltaE_00, err_geo, T_arr, delta_form.
• Service-side: T_proc (processing latency), rho (arrival/service ratio), P99 quantile.

III. Axioms P215- (Scenario Baseline)*

• P215-1 Linear-first: perform all calibration, measurements, and decisions on y_lin.
• P215-2 Two-form parallel: for any use case involving propagation time, compute both T_arr forms and record delta_form.
• P215-3 Unified time base: compare/window on tau_mono; publish on ts with offset/skew/J.
• P215-4 Traceable: write artifacts, parameters, and intermediates to audit_report with hash_sha256(blob) and signature.
• P215-5 Policy-driven SLO: decide using high quantiles (P95/P99), not means.

IV. Minimal Equations S215-*

• S215-1 Imaging error
• MSE = ( 1 / |Omega| ) * ( ∑_{r ∈ Omega} ( y_lin(r) - y_ref(r) )^2 )
• PSNR = 10 * log10( MAX^2 / MSE )。
• S215-2 Time/path gating & arrival harmonization
  A_path(y) = ( ∫ w(t) * y(t) dt ), where w(t) is the gate;
  delta_form = | ( 1 / c_ref ) * ( ∫ n_eff d ell ) - ( ∫ ( n_eff / c_ref ) d ell ) |.
• S215-3 Runtime stability
  rho = lambda_arrival / mu_service, W_q.P99 ≤ tol_queue, T_proc.P99 ≤ tol_latency.

V. Use Case 1: Offline Batch Pipeline (Multispectral Camera) — M150-1

1. Device/mode binding: register Device=MSI, Mode={ bandset, exposure, gain }; load response curves and bandpass (Chapter 3).
2. Linearization & radiometry: invert response, black/gain correction; validate unit, dim (Chapter 4).
3. Optics & resolution: compute MTF(f) via slanted-edge or point-source; derive MTF50, MTF_area (Chapter 5).
4. Sampling & reconstruction: mosaic/multi-resolution reconstruction of y_lin; cross-scale interpolation if needed (Chapter 6).
5. Noise modeling & denoising: estimate sigma_read, k_shot, NPS(f); choose structure-preserving denoisers (Chapter 7).
6. Flat/dark & FPN: build PRNU, DSNU, pixel health map; mask bad pixels (Chapter 8).
7. Geometry & registration: calibrate H and correct distortion; report err_geo (Chapter 9).
8. Color management: map multispectral to target color space; evaluate DeltaE_00.P95 (Chapter 10).
9. Computational imaging: deconvolution or SR for low-contrast/defocused samples (Chapter 12).
10. QC & freeze: aggregate SLI.img.*; judge against SLO.*; emit manifest.imaging.batch and audit_report.

• Typical contracts:
  • MTF50 ≥ 0.35 * f_Nyq; PRNU ≤ 0.015; DeltaE_00.P95 ≤ 3.0; err_geo ≤ 0.3 px.
  • repeatability ≤ tol_repeat; delta_form ≤ tol_Tarr (if arrival-time is involved).
• Deliverables: D_img.clean, SLI.img.*, manifest.imaging.batch, audit_report, full-pipeline TraceID.

VI. Use Case 2: Online Real-Time Imaging Service (Mobile/Edge) — M150-2

1. Streaming graph: on G=(V,E), build nodes ingest → linearize → denoise → demosaic → color → geometry → hdr → qc → publish (Methods.Cleaning v1.0, Ch. 11).
2. Time-base alignment: align on tau_mono; record offset/skew/J; bind windowed metrics to frame ts.
3. Low-latency design: single-pass imaging + QC; target T_proc.P99 ≤ 33 ms, rho ≤ rho_max.
4. Online QC: over sliding Delta_t, track SLI.img.mtf50.P95, SLI.img.deltaE00.P95, SLI.img.nps_band.
5. Backpressure loop: if W_q exceeds thresholds, downscale or disable expensive branches (SR/deconv); maintain drop_rate ≤ tol_drop.
6. Compliance & release: check_slo → emit_qc_manifest → freeze_release; divert exceptions to quarantine.
   • Typical SLOs: T_proc.P99 ≤ 33 ms; drop_rate ≤ 1%; SLI.img.mtf50.P95 ≥ 0.30 * f_Nyq; SLI.img.deltaE00.P95 ≤ 4.0.
   • Deliverables: runtime telemetry SLI.svc.* and SLI.img.*, plus manifest.imaging.realtime.

VII. Use Case 3: Event Camera + ToF Fusion (Time/Path Gating) — M150-3

• Acquisition & sync: align event stream E(t) and ToF depth D(t) on tau_mono.
• Path gating: integrate E(t) with w(t) = rect( ( t - t0 ) / Δt ) to form A_path(E); adjust t0, Δt using D(t).
• Two-form arrival: compute both T_arr forms; produce delta_form and set tol_Tarr.
• Spatial registration & fusion: estimate cross-modal H_evt↔tof; output err_geo and fused image y_fuse.
• QC & publish: measure motion_blur_index, ghosting, S_sync = max(|offset|,|skew|,J); upon compliance, emit manifest.imaging.fused.
  Typical contracts: S_sync ≤ tol_sync; delta_form ≤ tol_Tarr; err_geo ≤ 0.5 px; ghosting ≤ tol_ghost.

VIII. Scenario-Specific Contracts & Assertions

• Batch: assert { mtf50_floor, prnu_max, dsnu_max, color_fidelity, geometry_acc }.
• Real-time: assert { latency_p99, drop_rate, mtf50_p95, nps_band_max }.
• Event/ToF: assert { sync_ok, arrival_consistency, fusion_geo_ok }.
• General: assert audit_reproducible and all check_dim(expr) pass.

IX. Reference Binding I150-*

1. Build & run
   • build_offline_pipeline(cfg) -> pipe
   • run_offline_batch(pipe, inputs) -> { D_img.clean, SLI, manifest, audit_report }
   • build_realtime_graph(topology, policy) -> G
   • run_realtime(G, stream) -> telemetry(SLI.svc.*, SLI.img.*)
2. Reusable operators (referencing earlier I*-*)
   • linearize_and_calibrate(y_raw, RefCond) -> y_lin (Chapter 4)
   • measure_mtf_slanted_edge(img, roi) -> { MTF(f), f50, area } (Chapter 5)
   • estimate_noise_psd(seq, roi) -> { NPS(f), sigma_read, k_shot, NPS_band } (Chapter 7)
   • compute_prnu_dsnu(flats, darks) -> { PRNU, DSNU, map_prnu, map_dsnu, dead_pixel_rate } (Chapter 8)
   • calibrate_geometry(pattern_imgs) -> { H, err_geo } (Chapter 9)
   • evaluate_color(chart_raw, illum, profile) -> { DeltaE00_stats, wb_error } (Chapter 10)
   • hdr_exposure_fusion(frames, weights) -> y_hdr (Chapter 11)
   • deconv_or_superres(img, psf, method) -> img' (Chapter 12)
   • gate_by_tof(E(t), D(t), t0, Δt) -> A_path(E) (Chapter 13)
   • arrival_time_consistency(gamma, n_eff, c_ref) -> { T_arr_const, T_arr_general, delta_form } (Methods.Cleaning v1.0, Ch. 6)
   • aggregate_sli(metrics) -> SLI_dict ; check_slo(sli, policy) -> { pass, violations[] } (Chapter 14)
   • freeze_release(artifacts, tag) -> manifest (Methods.Cleaning v1.0, Ch. 10)

X. Cross-References

• Devices & modes: Chapter 3; Radiometric harmonization: Chapter 4; Optical resolution: Chapter 5; Sampling: Chapter 6.
• Noise & FPN: Chapters 7–8; Geometry & registration: Chapter 9; Color: Chapter 10; HDR: Chapter 11; Computational imaging: Chapter 12; Time/path gating: Chapter 13; Quality & audit: Chapter 14.
• Streaming & backpressure: Methods.Cleaning v1.0, Chapter 11; arrival two forms & release freeze: Methods.Cleaning v1.0, Chapters 6 and 10.

XI. Quality Metrics & Risk Control (Scenario Coupling)

• Offline: full SLI.img.* statistics with re-measure repeatability audits; failures route to rework queues.
• Real-time: monitor T_proc.P99, drop_rate, rho; trigger degradation and circuit-breaker policies; auto-rollback to safe configs.
• Event/ToF: focus on S_sync, delta_form, err_geo, and ghosting; on breach, isolate and mark data untrusted.

Summary
This chapter delivers three actionable blueprints—with P/S/M/I bindings—for a closed loop from device to publication. Thresholds and policies come from policy cards and SLO configs and are tunable by scenario. All outputs are auditable, traceable, and repeatable.