heyburns-PortraitUtils/docs/StripSolidBorders.md

StripSolidBorders

StripSolidBorders repeatedly trims thick, almost-flat colour frames that surround archive scans and scrapbook imports. The node keeps shaving edges while they stay uniform, making it ideal for cleaning large batches of legacy material.

---

Inputs

• image – Single RGB tensor; batching is not yet supported.
• max_border_percent – Hard stop for how much of the width/height can be removed on any edge. Lower it to protect content. Expressed as percent of the current dimension.
• variance_threshold – Maximum per-edge colour variance allowed before the algorithm stops stripping. Drop it when borders are extremely flat; raise it for noisier frames.
• colour_delta_threshold – Largest per-channel deviation (0–1 range) compared to the running edge mean. Increase when borders include gentle gradients.
• min_band_px – Minimum consecutive pixels per pass that must qualify as border before they are removed.
• passes – Number of full sweeps. Extra passes help chew through multi-stage mats or alternating light/dark stripes.
• mean_follow – Smoothing factor for the running colour estimate. Higher values let the baseline colour drift slowly as the strip progresses inward.
• min_uniform_fraction – Required portion of pixels in each sampled row/column that fall within the threshold. Relax this for grainy scans.
• dominant_bins – Histogram granularity (per channel) used for the fallback dominance test. Raising the count tolerates more subtle variations.
• edge_retry_allowance – Number of “outlier” rows/columns allowed before the pass halts. Increase when a few bright scratches interrupt the border.

---

Outputs

• image – Cropped tensor with the surrounding frame removed.
• trim_left, trim_top, trim_right, trim_bottom – Total pixel counts removed from each edge.
• detected – True when at least one edge was stripped; otherwise the input image is returned unchanged and counts are zero.

---

Usage Notes

• Feed the node individual images or pre-sliced batches—the implementation currently expects B=1.
• Start with the defaults. If borders remain on one side, increase passes or relax both variance_threshold and colour_delta_threshold.
• Because edges are evaluated independently, you can deliberately stack it after StripBottomBanner to remove bottom tickers first and then clean the remaining sides.

---

Troubleshooting

• Nothing trimmed – Either the edge noise exceeded variance_threshold, or the border exceeded max_border_percent. Try raising both thresholds or lowering min_uniform_fraction.
• Content lost – Reduce passes or tighten max_border_percent so the crop stops earlier. Inspect the returned trim counts to see which edge over-trimmed.
• Alternating stripes survive – Borders with multiple colours may need a higher edge_retry_allowance and mean_follow so the detector can adapt between stripes.