MUSA RIPENESS · 3-STATE · RGB DOMAIN

A statistical vision algorithm from 2020. No machine learning. Runs entirely in your browser.

CALIBRATED 2020 NO SERVER NO AI MODEL v0.1

[ ABSTRACT ] TECHNICAL NOTE · v0.1

A training-free classifier that labels a banana as unripe, ripe, or overripe from the mean RGB colour of the segmented fruit alone. Conceived as 2020 coursework and rebuilt here to run entirely in the browser, it is accurate under controlled lighting and fails in well-understood ways outside it. The full technical note covers the method, reference measurements, limitations, and references.

[ ACQUISITION ] DROP · CLICK · CLASSIFY

Drop a banana photo here

or click to browse

JPG · PNG · WEBP · MAX 8MB

TEST SAMPLES click any to classify

CH.A · SOURCE —

CH.B · SEGMENTED · THR=190 —

CLASSIFICATION OUTPUT

—

[ ANALYSIS ] SIX READOUTS

01 · COLOR DECOMPOSITION

Mean RGB across detected fruit area

—

Chlorophyll absorbs red and blue, reflects green. As it breaks into carotenoids, red and yellow rise, blue drops. Further breakdown into melanoidins drops everything toward dark brown.

02 · INTENSITY

RGB Average

—

OF 255

Threshold: 100
Status: —

03 · LOGIC PATH

Active rule

1 G > R > B → Unripe

2 R > G > B & R≥100 → Ripe

3 avg<80 & R<100 → Overripe

04 · REFERENCE

Against 2020 baseline data

CATEGORY	R̄	Ḡ	B̄	AVG
Unripe	44	162	128	111
Ripe	178	138	31	116
Overripe	73	52	57	61
YOUR IMAGE	—	—	—	—

05 · ORIGIN

Where the three conditions came from

Born from a Digital Image Processing assignment at ITN Malang, 2020. One week, during the Eid al-Adha holiday. No reference papers, no Google, no AI.

"Unripe bananas must be green-dominant, ripe ones yellow, rotten ones dark. So just compute the mean RGB and compare the ordering."

A decade later, the intuition still holds. Not perfectly accurate. Doesn't generalize to every condition. But enough, and transparent.

06 · LIMITATIONS

What this algorithm cannot do

Non-white backgroundsThreshold-190 segmentation assumes a white studio background. Market, tree, or in-hand photos break it.

Poor lightingYellow, neon, or shadowed light shifts RGB values systematically and produces wrong categories.

Non-Cavendish varietiesKepok, Raja, or Mas bananas have different natural color ranges. Thresholds tuned on Cavendish are biased.

Partial ripenessA half-yellow half-green banana is averaged into one category, losing spatial information.

No confidence scoreNo way to separate 'definitely ripe' from 'maybe ripe'. Every output is binary.

Not object detectionThe algorithm doesn't know if the photo is a banana. An apple or a sandal still gets classified by its RGB.