Counting Particles Without Fiji: A Free Browser Tool and How It Works

Sep 27, 2026 · @Frank Lloyd

Why this exists

You can count particles in a microscopy image without installing anything: this free tool runs the core of Fiji’s “Analyze Particles” workflow inside your web browser.

Counting cells, beads, vesicles, colonies or debris is one of the most common image tasks in any lab. The standard tool is Fiji (ImageJ). It is excellent, but it is not always available: locked-down work laptops, Chromebooks, tablets, a borrowed computer at a community lab, or a Java install that simply won’t cooperate.

This tool covers the everyday case: load an image, remove uneven background, threshold, count, and export a table. It is built for beginners and small labs who need a quick, transparent count, not a replacement for a full image-analysis suite.

What the tool does

The Particle Counter takes one image and returns a numbered overlay, summary statistics and a downloadable table of every particle.

• Works on fluorescence (bright particles) and brightfield or stained images (dark particles).
• Removes uneven illumination before thresholding.
• Picks a threshold automatically (Otsu’s method), or lets you set it by hand.
• Filters by size and can drop particles cut off by the image edge.
• Reports area and diameter in pixels, or in µm if you enter your pixel size.
• Counts inside a rectangular or circular region of interest (ROI).
• Exports a CSV (with your settings recorded at the bottom) and a PNG of the overlay.

Your images stay on your computer. All processing happens in the browser tab. Nothing is uploaded to a server, which matters for unpublished data, patient-derived samples or anything under an NDA.

How to use it

A first count takes about a minute. Open the tool here: Particle Counter.

1. Load an image. Drop a PNG or JPG onto the dark panel, or click Try a sample image to practice. TIFF files usually need converting to PNG first.
2. Say whether particles are bright or dark. Fluorescence is usually bright on dark. Stained or brightfield cells are usually dark on light.
3. Set the background radius. Start a little larger than the radius of your biggest particle. The tool warns you if it is too small.
4. Check the Mask view. Particles should be white and background black. If not, untick Automatic and move the threshold slider.
5. Set a minimum area. 8 px removes single-pixel noise. Raise it to ignore debris.
6. Enter µm per pixel (optional) to get real units and particles per mm².
7. Draw an ROI (optional) to count one area only. Particles are counted by their center point.
8. Download the CSV and overlay PNG. Keep both with your raw image so the count can be checked later.

How it works under the hood

The tool runs six steps, the same core sequence as Fiji’s Subtract Background → Threshold → Analyze Particles. Knowing them makes it much easier to fix a bad count.

1. Convert to grayscale. Color is collapsed to one brightness value per pixel, weighted the way the eye sees it (green counts most). For dark particles the image is inverted, so every later step can treat particles as bright.
2. Subtract the background. Microscope images are rarely evenly lit; one corner is often brighter than another. The tool estimates the background with a morphological opening: it slides a square window across the image, takes the local minimum, then the local maximum of that result. Anything smaller than the window (your particles) disappears from this estimate, while slow shading survives. Subtracting it leaves particles on a flat, near-zero background. This is the same idea as Fiji’s “rolling ball,” which is why the radius must be larger than your particles.
3. Smooth. A small averaging filter (3 × 3 pixels, run once or three times) blurs away pixel-level noise so particles don’t break into fragments.
4. Threshold. Every pixel is labeled particle or background by one cutoff. Otsu’s method picks the cutoff automatically by testing all 256 gray levels and choosing the one that best separates the image into two groups. It works well when particles and background are clearly different; you can override it.
5. Find connected particles. The tool scans the black-and-white mask and groups touching white pixels (including diagonal neighbors) into individual particles. For each one it records area, center, and mean original brightness.
6. Filter and measure. Particles below the minimum area, above the maximum, or touching the edge (if selected) are dropped. Diameter is reported as the diameter of a circle with the same area. Pixel values convert to µm when you give the pixel size.

Large images are analyzed at a maximum of 2000 pixels on the long side to keep the browser responsive. Areas and positions are converted back to the original image’s pixels in the results.

Choosing parameters

Most bad counts come from three settings: polarity, background radius and threshold. Use the Mask view to diagnose, then match what you see below.

What you see	Likely cause	Fix
Zero or very few particles	Polarity is wrong	Switch Bright / Dark
Large particles hollow or missing	Background radius too small	Raise the radius above the largest particle’s radius
Big blobs where shading was	Background radius too large, or off	Lower the radius, or turn it on
Particles split into fragments	Too much noise	Set smoothing to Strong, or lower the threshold
Specks counted as particles	Noise above threshold	Raise the minimum area
Neighbors merged into one	Touching particles	Raise the threshold slightly; see Limitations
Half-particles along the border	Edge cut-off	Tick Exclude particles touching the edge

Once the settings work for one image, use the same settings for every image in the experiment. Changing them between conditions can create a difference that isn’t biological. The CSV records the settings at the bottom so you can reproduce them.

Limitations and how to validate

This is a first-pass counting tool, not a validated measurement instrument. Check it against a trusted method before relying on it.

Known limitations:

• Touching particles count as one. There is no watershed splitting yet. In the built-in sample (140 randomly placed cells) it counts 123, because overlapping cells merge. Dense samples will be undercounted.
• One channel only. Color images are converted to gray; there is no per-channel analysis.
• 8-bit display data. The browser reads images as 8-bit. 16-bit TIFFs lose dynamic range when converted, and some won’t open at all.
• Square window. Background subtraction uses a square window, not Fiji’s ball. Results are close but not identical.
• No shape measures. Circularity and perimeter are not reported, because pixel-edge perimeters are unreliable at small sizes.

A simple validation you can do in an afternoon:

1. Pick 5–10 representative images, including a crowded one.
2. Count each by hand, or in Fiji if you can reach a computer that has it.
3. Count the same images here with one fixed set of settings.
4. Compare the counts image by image. Consistent differences are fine if they are small and the same direction; large or random differences mean the settings, or this tool, don’t suit your images.

Report the tool, its settings and your validation in your methods section.

How AI was used to build it

The tool was built with an AI assistant (Claude), starting from a Python pipeline that reproduced Fiji’s workflow with scikit-image and scipy. The practical lesson for DIY builders: AI is fastest when you already know what the method should do.

• Start from a method you understand. The Python version came first, checked against Fiji’s steps. Porting a known pipeline is far safer than asking for “a particle counter” from scratch.
• Ask for the reasoning, not just code. Understanding why the background radius must exceed particle size made the tool’s warnings possible.
• Test on data with a known answer. The core code was checked on synthetic images with a known number of particles and known areas before any real image was used.
• Keep it simple to host. The final tool is one HTML file with no server, which costs nothing to run and keeps images private.

Planned next: watershed splitting of touching particles, and batch processing of a folder of images with one settings file. Tell us which you would use first through the Submit Articles page.