Count Objects in Image Online

How the Object Counter works

How to count objects in image online

Counting by threshold works whenever the objects are noticeably lighter or darker than what they sit on: dark coins on a white sheet, pale cells on a stained slide, seeds on a dark cloth. The image is converted to greyscale and lightly blurred to suppress sensor noise, then every pixel is classed as object or background by a cut-off brightness.

Automatic picks that cut-off with Otsu's method, which searches every possible level and chooses the one that best separates the two brightness populations in the picture. It suits most well-lit photos on a plain background. Manual lets you set the level yourself with the slider, which is the answer when the automatic choice grabs shadows or misses faint objects; the level actually used is reported in the stats so you can start from the automatic value and nudge it. The tool looks at the border of the picture to decide which side of the cut-off is background, so it does not matter whether your objects are lighter or darker than the surface.

After thresholding, a small opening pass removes single-pixel grit, and anything smaller than the minimum object size is ignored. Raise that value when dust, texture or reflections are being counted; lower it for genuinely tiny objects like pollen or fine beads.

Split touching objects deals with the classic problem of two coins or cells overlapping into one blob. It computes how far each object pixel is from the nearest background pixel, so the centre of each round object becomes a peak, uses those peaks as seeds, and runs a watershed that grows each seed outwards until it meets its neighbour. Round or oval objects separate cleanly; long thin objects or very irregular shapes can be over-split, in which case turn it off and accept that touching pairs count as one.

The result image tints every counted object with its own colour and prints the total in the corner, and the stats give the smallest, largest and median object area in pixels. A quick scan for a blob that is much larger than the median usually reveals the pair that did not split, and a cluster of tiny areas points to the minimum size being too low. Images over 2000 pixels on the long side are scaled down before counting.

Related tools