Image Dehazing

Module: Computer Vision | Difficulty: Intermediate

Atmospheric Scattering Model

where is scene radiance, is transmission, is atmospheric light.

Dark Channel Prior

Transmission Map

Scene Radiance Recovery

import cv2
import numpy as np

def dark_channel(img, size=15):
    b, g, r = cv2.split(img)
    dc = cv2.min(cv2.min(r, g), b)
    kernel = cv2.getStructuringElement(cv2.MORPH_RECT, (size, size))
    return cv2.erode(dc, kernel)

def atmospheric_light(img, dark_ch):
    _, _, _, top_left = cv2.minMaxLoc(dark_ch)
    return img[top_left[1], top_left[0]]

def transmission_map(img, A, size=15, omega=0.95):
    return 1 - omega * dark_channel(img / A, size)

def dehaze(img, A, t, t_min=0.1):
    t = np.clip(t, t_min, 1)
    return np.clip((img - A) / t + A, 0, 255).astype(np.uint8)

Key Takeaways

• Dark channel prior is effective for natural outdoor haze
• Transmission map estimation is key to dehazing quality
• Deep methods learn end-to-end dehazing without priors