imreducehaze

Reduce atmospheric haze

Syntax

[J,T,L] = imreducehaze(I)

[___] = imreducehaze(I,amount)

[___] = imreducehaze(___,Name,Value)

Description

[J,T,L] = imreducehaze(I) reduces atmospheric haze in color or grayscale image I. The function returns the dehazed image J, an estimate T of the haze thickness at each pixel, and the estimated atmospheric light L.

[___] = imreducehaze(I,amount) additionally specifies the amount of haze to remove.

[___] = imreducehaze(___,Name,Value) changes the behavior of the dehazing algorithm using name-value pairs.

Examples

Reduce Haze Using Default Parameters

Read a hazy image into the workspace.

A = imread('foggysf1.jpg');

Reduce the haze and display the result next to the original image in a montage.

B = imreducehaze(A);
montage({A,B})
title("Hazy Image (Left) vs. Reduced Haze Image (Right)")

Reduce Haze Using approxdcp Contrast Stretching

Read a hazy image into the workspace.

A = imread('foggysf2.jpg');

Reduce 90% of the haze using the approxdcp method.

B = imreducehaze(A,0.9,'method','approxdcp');

Display in a montage the original hazy image and the image with reduced haze.

montage({A,B})

Estimate Haze Thickness and Image Depth

Read hazy image into the workspace.

A = imread('foggyroad.jpg');

Reduce haze in the image using default parameter values.

[~, T] = imreducehaze(A);

Display the result along side the original image.

figure, imshowpair(A, T, 'montage')

The haze thickness provides a rough approximation of the depth of the scene, defined up to an unknown multiplication factor. Add eps to avoid log(0).

D = -log(1-T+eps);

For display purposes, scale the depth so that it is in [0,1].

D = mat2gray(D);

Display the original image next to the estimated depth in false color.

figure
subplot(1,2,1)
imshow(A), title('Hazy image')
subplot(1,2,2)
imshow(D), title('Depth estimate')
colormap(gca, hot(256))

Input Arguments

I — Hazy image

RGB image | grayscale image

Hazy image, specified as an RGB or grayscale image.

Data Types: single | double | uint8 | uint16

amount — Amount of haze to remove

1 (default) | number in the range [0,1]

Amount of haze to remove, specified as a number in the range [0,1]. When the value is 1, imreducehaze reduces the maximum amount of haze. When the value is 0, imreducehaze does not reduce haze and the input image is unchanged. Larger values can cause more severe color distortion.

Data Types: single | double | int8 | int16 | int32 | int64 | uint8 | uint16 | uint32 | uint64

Name-Value Pair Arguments

Specify optional comma-separated pairs of Name,Value arguments. Name is the argument name and Value is the corresponding value. Name must appear inside quotes. You can specify several name and value pair arguments in any order as Name1,Value1,...,NameN,ValueN.

Example: B = imreducehaze(A,0.9,'method','approxdcp');

'Method' — Technique used to reduce haze

'simpledcp' (default) | 'approxdcp'

Technique used to reduce haze, specified as the comma-separated pair consisting of 'Method' and one of these values:

• 'simpledcp' — Simple dark channel prior method [2]. This method uses a per-pixel dark channel to estimate haze and quadtree decomposition to estimate the atmospheric light.

• 'approxdcp' — Approximate dark channel prior method [1]. This method uses both per-pixel and spatial blocks when computing the dark channel and does not use quadtree decomposition.

For more information, see Algorithms.

Data Types: char | string

'AtmosphericLight' — Maximum value to be treated as haze

1-by-3 numeric vector | numeric scalar

Maximum value to be treated as haze, specified as the comma-separated pair consisting of 'AtmosphericLight' and a 1-by-3 numeric vector for RGB images or a numeric scalar for grayscale images. Values must be in the range [0, 1]. Atmospheric light values greater than 0.5 tend to give better results.

If you do not specify AtmosphericLight, then the imreduzehaze function estimates a value depending on the value of 'method'.

Data Types: double

'ContrastEnhancement' — Contrast enhancement technique

'global' (default) | 'boost' | 'none'

Contrast enhancement technique, specified as the comma-separated pair consisting of 'ContrastEnhancement' and 'global', 'boost', or 'none'.

Data Types: char | string

'BoostAmount' — Amount of per-pixel gain

0.1 (default) | number in the range [0, 1]

Amount of per-pixel gain to apply as postprocessing, specified as the comma-separated pair consisting of 'BoostAmount' and a number in the range [0, 1]. This argument is only supported if ContrastEnhancement is specified as 'boost'.

Data Types: double

Output Arguments

J — Dehazed image

numeric array

Dehazed image, returned as numeric array of the same size as the input hazy image I.

T — Haze thickness

numeric array

Haze thickness estimated at each pixel, returned as a numeric array.

L — Estimated atmospheric light

numeric array

Estimated atmospheric light, returned as a numeric array. L represents the value of the brightest nonspecular haze.

Algorithms

The model to describe a hazy image I is

I(x) = J(x)T(x) + L(1-T(x))

I is the observed intensity, J is the scene radiance, L is atmospheric light, and T is a transmission map describing the portion of light that reaches the camera.

Dehazing algorithms recover the scene radiance (dehazed image) J from an estimation of the transmission map and atmospheric light according to:

J(x) = (I(x)-A)/(max(t(x),t0)) + A

imreducehaze uses two different dehazing algorithms, simpledcp and approxdcp. These methods both rely on a dark channel prior, which is based on the observation that unhazy images of outdoor scenes usually contain some pixels that have low signal in one or more color channels. The methods differ in how they estimate the dark channel prior and atmospheric light.

The dehazing algorithms in imreducehaze follow five steps:

1. Estimate the atmospheric light L using a dark channel prior.

2. Estimate the transmission map T.

3. Refine the estimated transmission map.

4. Restore the image.

5. Perform optional contrast enhancement.

References

[1] He, Kaiming. "Single Image Haze Removal Using Dark Channel Prior." Thesis, The Chinese University of Hong Kong. 2011.

[2] Dubok, et al. "Single Image Dehazing with Image Entropy and Information Fidelity." ICIP. 2014, pp. 4037–4041.