11.9 Image processing with OpenCV

Source: Ideas from Computational Physics (Landau, Páez, Bordeianu), §11.8. Rewritten in original wording — no textbook OpenCV listing dumps. Continues 11.8b. Next: 11.9b.

Problem: separate ripe strawberries from green ones (or roast levels of coffee beans) using images.

OpenCV (opencv-python) is a computer-vision library with imaging and ML helpers. Install locally with pip install opencv-python (or the contrib build). Not available in this sandbox — labs use pure-Python histogram / difference arithmetic.

RGB pixels

Images are grids of pixels. Each pixel stores R, G, B channel strengths. One byte per channel ⇒ $2^8=256$ levels (0–255). Distinct RGB triples:

$$256\times 256\times 256 = 16{,}777{,}216.$$

OpenCV often stores channels as BGR order when reading files — channel index $0$ is blue, $1$ green, $2$ red in calcHist examples.

Histograms for ripeness

calcHist counts how many pixels fall in each of 256 tone bins per channel. Ripe fruit tends to pile counts into red high bins; unripe fruit shifts weight toward green. Those histogram shapes become features for a classifier (or a simple threshold on red/green ratios).

# rgb_hist_sketch.py — structure (local OpenCV)
# img = cv.imread("ripe.jpg")
# for ch, color in enumerate(("b", "g", "r")):
#     hist = cv.calcHist([img], [ch], None, [256], [0, 256])
#     # plot hist

Background subtraction

For video (exoplanet transits, lab pistons, smoke): keep what changes. createBackgroundSubtractorMOG2() builds a statistical background model and returns a foreground mask. Simpler pedagogy: absolute difference between consecutive frames, then threshold.

Takeaway

Tool	Use
RGB histogram	Color / ripeness signatures
Channel order	Remember BGR vs RGB
Frame differencing / MOG2	Drop static background

Quiz

OpenCV / images

1 / 4

Levels per RGB channel with one byte are…

Pick an answer to see if you are right.

Next: 11.9b OpenCV labs.