Member Avatar for senexom

Hey does anyone know of a way to crop images automatically while still focusing on the subject in the photo? I have a gallery of pictures and I would like to make all the thumbnails square shaped, just thought maybe someone knows if there is such a thing as smart crop or something of that nature. Perhaps let the user crop their own photos, how would one go about that w/ php + GD.

  • 2 Contributors
  • 3 Replies
  • 104 Views
  • 4 Days Discussion Span
  • Latest Post Latest Post by senexom

Dani AI

Generated

A practical way to get subject-aware square thumbnails (the goal described) is a two-step pipeline: 1) find a region of interest (ROI) that contains the subject, then 2) crop that ROI into a square and resize. For people, use face detection; for general photos use an algorithmic saliency/edge/entropy scan. The layering/overlay idea mentioned by is fragile; prefer an explicit crop (client-side canvas UI for manual tweaks, or server-side detection for automation).

Face-detection route: run a detector (Haar/DNN) to get face boxes, pick the largest or center-weighted box, expand it 10–30% to avoid tight cuts, then constrain to image bounds and convert that bbox to a square before resizing. In a PHP setup this is often easiest by calling a small CLI microservice (OpenCV) or using a PHP OpenCV extension, then doing the final image ops with GD or ImageMagick. See OpenCV for detectors (OpenCV docs) and the PHP image functions (PHP GD manual). ImageMagick's -gravity/-crop are convenient for server-side cropping.

Saliency / edge / entropy route (no faces): compute a grayscale edge or entropy map, build an integral image and evaluate candidate square windows quickly to find the one with the highest summed saliency—this targets high-contrast, textured, or information-dense areas. High-level pseudocode:

# saliency-based square crop (high-level)
saliency = edge_magnitude(grayscale(img))
integral = integral_image(saliency)
best_score, best_box = 0, null
for size in candidate_square_sizes:
  for x,y over image grid step S:
    score = sum_region(integral, x,y,size,size)
    if score > best_score:
      best_score, best_box = score, (x,y,size)
crop = expand_to_fixed_square(best_box, target_size)
resize_and_save(crop)

Practical notes: handle multiple faces by heuristics (largest face, center bias, or average); allow a manual override (client-side crop via canvas) for tricky shots; cache thumbnails and do heavy processing in background jobs. For very large images GD can hit memory limits—prefer ImageMagick or libvips for production. Be explicit about margins so heads/limbs aren't cropped off, and fall back to a center-crop when detection fails.

Recommended Answers

All 3 Replies

Member Avatar for Member #36984

Wow, that's pretty complex PHP!

Although it is miles out of my knowledge, I do have an idea:

Maybe you could have parts of the pictures hidden behind some kind of layers to match your page.

Can I emphasise again that i really don't know... lol :lol: -- I only have some basic knowledge of PHP!

Cheers

Member Avatar for senexom

I'd prefer to stay away from using layers to cover images, mostly because of all the incompatibilities between browsers, but in a pinch I have done some funky stuff with DHTML and CSS though.

Be a part of the DaniWeb community

We're a friendly, industry-focused community of developers, IT pros, digital marketers, and technology enthusiasts meeting, networking, learning, and sharing knowledge.