Drop images here
PNG, JPG, WEBP, GIF — up to 20 images at once
Choose files
How it works
The Math
Image → grayscale → 3×3 Laplacian kernel (2nd-order derivative, kernel: center×4 minus 4 neighbors). Variance of those values = sharpness score. High variance = many strong edges = sharp.
Sharp ≥ 500
Lots of edges, high-intensity pixel transitions. Clear photo, crisp text, detailed scene. Reliable for machine vision and analysis tasks.
Okay 100–499
Moderate edge content. Usable but imperfect — could be intentional shallow DoF, mild motion blur, or compressed image.
Blurry < 100
Few edges, flat Laplacian response. Camera shake, defocus, or heavy filtering. Threshold of 100 is the canonical research benchmark — see paper below.
Score 0–100
Raw variance mapped to 0–100 via log₁₀ scale, since natural images span 0 to 100,000+. The heatmap visualizes exactly where the Laplacian response is strongest.
Edge Heatmap
Purple = low Laplacian response (flat/blurry regions). Cyan/bright = high response (sharp edges, fine detail). Instantly reveals where focus landed in your image.
📄 FOUNDATIONAL PAPER
Diatom Autofocusing in Brightfield Microscopy: A Comparative Study
Proceedings of the 15th International Conference on Pattern Recognition (ICPR 2000), pp. 3318–3321. IEEE Computer Society. DOI: 10.1109/ICPR.2000.903548
Originally developed for autofocusing microscopes used to identify diatoms (single-cell silica algae), this 2000 paper established Variance of Laplacian as a reliable, fast focus measure. The authors showed it outperformed prior methods including Tenengrad on speed, sharpness accuracy, and noise robustness. The threshold of 100 for blur classification emerged from this lineage of research. Today, the exact same algorithm runs in browsers, phone cameras, and production CV pipelines worldwide.