Image Detection Stage Reference — MediaCleaner Pro

Documentation Index

Fetch the complete documentation index at: https://mintlify.com/xcoder-es/media-cleaner-pro/llms.txt

Use this file to discover all available pages before exploring further.

​

Stage 1: Exact Duplicate Removal

let is_duplicate = seen_hashes.contains(&meta.sha256);

​

Stage 2: Perceptual Duplicate Removal

let prefix = (dhash >> 48) as u16;
// Checks buckets: prefix.wrapping_sub(1), prefix, prefix.wrapping_add(1)
for (existing_path, existing_hash) in &candidates {
    let distance = hamming_distance(dhash, *existing_hash);
    if distance <= self.threshold {
        duplicates.push(existing_path.clone());
    }
}

​

Stage 3: Tiny Image Detection

let is_tiny = meta.width < min_width || meta.height < min_height;

​

Stage 4: Icon Detection

1. Square aspect ratio — the width-to-height ratio falls between 0.9 and 1.1 (inclusive).
2. Small size — both dimensions are at or below 256 px.

let aspect = meta.width as f32 / meta.height.max(1) as f32;
let is_square = (0.9..=1.1).contains(&aspect);
let is_small = meta.width <= 256 && meta.height <= 256;
let is_icon = is_square && is_small;

​

Stage 5: Thumbnail Detection

1. Name-based match — the filename (lowercased) contains any of: thumb, thumbnail, small, mini, preview, tn_
2. Size-based match — width ≤ 320 px or height ≤ 240 px, and the image is not taller-than-wide by a 2:1 margin (to avoid catching valid portrait photos at small sizes)

let has_thumb_name = thumb_patterns.iter().any(|p| name_lower.contains(p));
let is_small = meta.width <= 320 || meta.height <= 240;
let is_thumbnail = has_thumb_name || (is_small && meta.width < meta.height * 2);

​

Stage 6: Screenshot Detection

1. Resolution match — the image’s width × height (or height × width for portrait captures) matches any of 16 preset screen resolutions:

   Resolution	Common device/use
   1920 × 1080	Full HD desktop/laptop
   1366 × 768	Common laptop
   1440 × 900	MacBook Air (older)
   1536 × 864	Windows scaled HD
   1280 × 720	HD / older display
   1600 × 900	HD+ laptop
   2560 × 1440	QHD / 27” display
   3840 × 2160	4K UHD
   1280 × 1024	Legacy 5:4 display
   1024 × 768	Legacy XGA / tablet
   1680 × 1050	WSXGA+ display
   1920 × 1200	WUXGA display
   1440 × 960	—
   2560 × 1600	MacBook Pro (older)
   2880 × 1800	MacBook Pro Retina
   3024 × 1964	MacBook Pro 14-inch

2. Name-based match — the filename (lowercased) contains screenshot, screen shot, or screencapture.

let is_common_res = common_resolutions.iter().any(|(w, h)| {
    (meta.width == *w && meta.height == *h) || (meta.width == *h && meta.height == *w)
});
let has_screenshot_name = name_lower.contains("screenshot")
    || name_lower.contains("screen shot")
    || name_lower.contains("screencapture");

​

Stage 7: Wallpaper Detection

1. Ultrawide aspect ratio — width ÷ height falls between 1.8 and 3.5 (inclusive). This covers 16:9 (1.78), 21:9 (2.33), and panoramic images.
2. 4K+ with wide aspect — width ≥ 3840 px or height ≥ 2160 px, and the aspect ratio is ≥ 1.5 (landscape orientation at 3:2 or wider).

let aspect = meta.width as f32 / meta.height.max(1) as f32;
let is_ultrawide = (1.8..=3.5).contains(&aspect);
let is_4k = meta.width >= 3840 || meta.height >= 2160;
let is_wallpaper = is_ultrawide || (is_4k && aspect >= 1.5);

​

Stage 8: Document Detection

1. Paper aspect ratio — the image’s aspect ratio (or its reciprocal, to handle portrait orientation) falls within ±0.15 of any of four standard paper ratios:

   Paper standard	Ratio
   A4 (ISO 216)	1.414 (√2)
   US Letter	1.294
   US Legal	1.545
   Square / CD	1.0

2. High resolution — width ≥ 1200 px, ensuring low-resolution letter-shaped images (small icons, avatars) are not mistakenly flagged.

let paper_ratios = [1.414, 1.294, 1.545, 1.0];
let is_paper_ratio = paper_ratios
    .iter()
    .any(|r| (aspect - *r).abs() < 0.15 || ((1.0 / aspect) - *r).abs() < 0.15);
let is_high_res = meta.width >= 1200;
let is_document = is_paper_ratio && is_high_res;

​

Stage 9: AI Classification

if aspect < 0.8                            { scores.push(("portrait",    0.7)); }
if aspect > 1.3 && mp > 2.0               { scores.push(("landscape",   0.6)); }
if mp > 5.0 && aspect > 1.0 && aspect < 1.5 { scores.push(("macro",    0.5)); }
if mp < 0.5                                { scores.push(("low_quality", 0.8)); }
if aspect > 2.5                            { scores.push(("panorama",    0.9)); }
if aspect > 0.95 && aspect < 1.05         { scores.push(("social",      0.6)); }

• A 2000×3000 px phone photo → portrait (aspect 0.67)
• A 4000×2500 px DSLR landscape → landscape (aspect 1.6, 10 MP)
• A 5000×3500 px detailed close-up → macro (17.5 MP, aspect 1.43)
• A 8000×2000 px stitched cityscape → panorama (aspect 4.0)
• A 1080×1080 px Instagram export → social (aspect 1.0)

​

Stage 10: Quality Ranking

let mp = (meta.width * meta.height) as f32 / 1_000_000.0;
let size_score = (meta.size_bytes as f32 / (mp * 500_000.0)).min(1.0) * 30.0;
let res_score  = mp.min(20.0) / 20.0 * 40.0;
let aspect_score = if meta.width > meta.height { 15.0 } else { 10.0 };
let total = (size_score + res_score + aspect_score + 15.0).min(100.0);

• A 20 MP RAW-exported JPEG with large file size → likely excellent or good
• A 2 MP photo saved at high compression → likely average
• A 0.3 MP image (below 0.5 MP threshold) → low resolution score, likely below_average or poor