Sharpness Scoring

Wed, 25 Mar 2026 00:00:00 +0000

Updated Sharpness Scoring in RawCull

This document describes how RawCull computes sharpness scores, what each parameter does, and a boundary-value test procedure for validating parameter behaviour. This applies to version 1.3.7 of RawCull

How the Scoring Pipeline Works

flowchart TD
 A([ARW File]) --> B[Thumbnail decode\nthumbnailMaxPixelSize]
 A --> C[Saliency detection\nApple Vision]

 B --> D[Gaussian pre-blur\npreBlurRadius × isoFactor × resFactor]
 D --> E[Laplacian Metal kernel\nfocusLaplacian]
 E --> F[Amplify\nenergyMultiplier]
 F --> G[Border suppression\nborderInsetFraction]
 G --> H[Render to Float32 buffer]

 H --> I[Full-frame samples\nborder-inset region]
 C -->|bounding box ≥ 3% area| J[Salient-region samples\nVision bounding box]
 H --> J

 I --> K[p95 winsorized tail score\nfull-frame f]
 J --> L[p95 winsorized tail score\nsubject s]

 K --> M{Fusion}
 L --> M
 C -->|area| M

 M --> N[finalScore]
 N --> O([Badge: score / maxScore × 100])

 style A fill:#2d2d2d,color:#fff
 style O fill:#2d2d2d,color:#fff

Each ARW file goes through the following stages in detail:

Sharpness Scoring

Wed, 25 Mar 2026 00:00:00 +0000

Sharpness Scoring

RawCull can automatically rank every image in a catalog by sharpness, so the sharpest frames rise to the top and out-of-focus or motion-blurred shots sink to the bottom. This page explains what the score means, how to run it, and how to interpret the results.

Overview

When you click Sharpness, RawCull analyses each RAW file using a multi-step pipeline:

A small thumbnail is extracted directly from the ARW file — fast, no full RAW decode needed.
Apple Vision locates the main subject in the frame and, when classification is enabled, identifies what that subject is (bird, person, animal, etc.). A salient region is accepted when it covers more than 3 % of the frame or when Vision’s confidence is ≥ 0.9 — the high-confidence path means a small but clearly identifiable subject (a bird on a stick, a distant raptor) is never discarded for being too small.
A Gaussian pre-blur smooths the image to suppress high-ISO noise before edge detection. The radius adapts automatically: it scales proportionally with √(ISO/400) to compensate for noise at high ISO values, and with √(imageWidth/512) to maintain consistent spatial-frequency selectivity on larger images. Both factors are capped at 3×.
A Laplacian edge-detection kernel measures per-pixel sharpness energy across the entire frame. The outer 4 % of each border is excluded from the full-frame score to prevent Gaussian edge artifacts from inflating the result.
The edge response is amplified, then scored with a robust tail score — the mean of values above the 95th percentile, clipped at the 99.5th — so a single sharp feather edge cannot carry a blurry frame.
Two scores are computed separately: one for the full frame (minus the border strip) and one for the salient subject region. They are fused as 25 % full-frame + 75 % subject region. If no salient region is found, the full-frame score is used directly. A small proportional bonus is applied for subjects that fill more of the frame.
All scores are normalised against the p90 of the catalog: that anchor becomes 100, and every other frame is expressed as a percentage of it. Using p90 rather than the maximum prevents a single noise spike from collapsing the rest of the scale.

Because scores are relative to your catalog, a score of 60 does not mean the image is soft — it means it is 60 % as sharp as the sharpest typical frame in that session. Opening a catalog of sharper images will recalibrate the scale.

Sharpness on RawCull

Sharpness Scoring

Updated Sharpness Scoring in RawCull

How the Scoring Pipeline Works

Sharpness Scoring

Sharpness Scoring

Overview