← all work classical-computer-vision

Dehazing

[![Python 3.10+](https://img.shields.io/badge/python-3.10+-3776AB?logo=python&logoColor=white)](https://www.python.org/)

Result

Sr	Scene	Hazy input	Dark channel prior	DCP + guided refine	CLAHE (contrast only)	Multi-scale Retinex	Gamma curve (control)	Oracle
1	whitewater raft · people, spray, close rock	11.5 dB	22.1 dB	23.1 dB	13.4 dB	8.8 dB	16.0 dB	50.7 dB
2	old street · receding to a vanishing point	12.7 dB	21.8 dB	23.9 dB	16.5 dB	13.2 dB	17.1 dB	50.8 dB
3	tropical island · sky-dominant	13.2 dB	22.7 dB	23.6 dB	13.7 dB	8.6 dB	18.8 dB	50.7 dB
4	motocross · near subject, shallow depth	10.8 dB	20.5 dB	21.2 dB	12.8 dB	9.1 dB	15.1 dB	50.7 dB

The finding

What it does

`mermaid flowchart LR A[Clear image
known exactly] --> B["Add haze
I = J·t + A·(1−t)"] B --> C[5 dehazing methods] B --> D[Oracle
handed the TRUE transmission] C --> E[Score vs the original
AND vs the true transmission map] D --> E E --> F[Did it remove haze,
or just add contrast?] `

Because the haze is generated, two things are known, not one:

That second row is what separates this from a contrast comparison. "Looks clearer" is satisfied by any histogram stretch; recovering t is only satisfied by actually inverting the scattering.

---

Elsewhere

Code and full results on GitHub (code private)

Every number is read from the file the run produced. github.com/hammasbuilds · kaggle.com/muhammadhammas13