nx_jpeg_sof.nx

nx_jpeg_sof.nx -- JPEG Start-of-Frame parser. Per ITU-T Rec. T.81 sec B.2.2 -- the SOF segment that establishes the image's dimensions, sample precision, and component descriptors. Baseline-DCT SOF0 (marker 0xC0) payload layout: Lf (2 bytes BE) -- segment length (already consumed by marker scanner; payload arrives without it) P (1 byte) -- sample precision (always 8 for baseline) Y (2 bytes BE) -- image height in pixels (rows) X (2 bytes BE) -- image width in pixels (cols) Nf (1 byte) -- component count (1=gray, 3=YCbCr, 4=CMYK) per component (3 bytes each): Ci (1 byte) -- component identifier (1..255) HiVi (1 byte) -- high4 = Hi sampling factor, low4 = Vi sampling factor Tqi (1 byte) -- quantization-table id (0..3) per T.81 sec B.2.2 Sampling factors are CRITICAL for chroma upsampling: e.g. 4:2:0 has Y component with H=2, V=2 and Cb/Cr with H=1, V=1 (luma sampled twice as densely in both axes). Progressive (SOF2 / 0xC2) payload format is identical to SOF0. Extended-sequential (SOF1 / 0xC1) is also identical. Only the scan-direction differs, which this parser doesn't care about. Lossless (SOF3 / 0xC3) and arithmetic-coded variants differ slightly -- not supported here. nx_safety_envelope: intended_use: "Frame-descriptor parsing for JPEG decode." sil_target: SIL1 evidence: [t81_section_b_2_2_canonical_basis, sealed_precision_enum, bounded_4_component_max] hazard_register: [bug-tape-sof-component-count-overflow, bug-tape-sof-malformed-precision, bug-tape-sof-zero-dimensions] residual_risk: "Caller must verify decoded Hi/Vi are in [1..4] before using as MCU sampling factor." verdict: NOT_YET_EVALUATED