Download as pdf or txt
Download as pdf or txt
You are on page 1of 8

International Journal of Advanced Science and Technology

Vol. 29, No. 05, (2020), pp. 10466 - 10473

ADAPTIVE AND HISTOGRAM EQUALISATION IMAGE


ENHANCEMENT METHODS FOR CORRECTING LOW
ILLUMINATION IMAGES

T. V. Hyma Lakshmi1, G. Satish2

1AssistantProfessor, Department of ECE, SRKR Engineering College, Bhimavaram, Andhra Pradesh,


India
2MTECH Student, Department of ECE, SRKR Engineering College, Bhimavaram, Andhra
Pradesh, India

Abstract:

In real world applications the need to extract the key features from low-illumination images, night
vision or abrupt weather conditions is a challenge. In order to fulfil this need the image enhancement
method is used in digital image processing to reduce the flaws in the image by brightening,
sharpening etc. Due to image enhancement the visibility and contrast of the image is improved. In this
paper, the colour correction for low illumination images is done using adaptive enhancement
technique and histogram equalisation method accompanied with fusion strategy. In this paper, the
RGB image is converted into HSV colour plane and V-plane is modified accordingly with the methods
to improve the brightness of the image. The entropy, PSNR, MSE of different images are calculated
and compared. The adaptive enhancement technique showed better results compared to histogram
equalisation. The enhanced images by these techniques are clear and bright when compared to input
image.
Keywords: Adaptive image enhancement, Histogram Equalisation, Brightness, Entropy, Peak Signal
to Noise Ratio, Mean Square Error.

I. Introduction
Digital image processing systems are widely used in many applications like remote sensing, medical
field, transmission and encoding, machine vision etc. The image enhancement is the process to
improve the display of the image or properties of image for further image analysis. Image acquisition
in adverse conditions like night-time, cloudy or smoky weather etc., results in many defects because
of weak reflection of light from object. The need to acquire clear images from these unfavourable
situations now became a challenge.
The aim of this project is to improve the entropy, PSNR of low illumination images. In this paper, the
low illumination images are corrected using adaptive enhancement and histogram equalization
techniques. The RGB space is converted to HSV space and value channel of the image are modified
using these methods to reduce uneven illumination and increase the adaptability of parameter settings.
The uneven illumination component is estimated using multiscale gamma function and distribution
profile is improved using adaptive enhancement method. Histogram equalisation method is used to
enhance the global contrast of image. After enhancement, image fusion strategy is used to extract the
details of the image. In both the methods the entropy PSNR, RMSE values of processed images are
increased compared to unprocessed images. Adaptive enhancement technique provides better results.
II. Related Work
The interest from digital image processing stems from two principal application areas: improvement
of pictorial representation for human interpretation, processing of image data for autonomous machine
perception [1]. Wang et al. proposed colour image correction method based on non-linear
transformation function based on light reflection model and multi-scale theory [2]. Traditional low-
illumination image enhancement algorithms include the gray transformation method and histogram
equalization method [3]. Huang et al. [4] proposed an adaptive gamma correction algorithm to
adaptively obtain gamma correction parameters based on a cumulative distribution probability
histogram. Jobson et al. proposed the single-scale retinex (SSR) algorithm [5] based on the retinex

ISSN: 2005-4238 IJAST 10466


Copyright ⓒ 2020 SERSC
International Journal of Advanced Science and Technology
Vol. 29, No. 05, (2020), pp. 10466 - 10473

illumination-reflection model established by Land et al. [6] , which later evolved into the multiscale
retinex (MSR) algorithm, the MSR algorithm with colour restoration (MSRCR) [7,8] , and the MSR
algorithm with chromaticity preservation (MSRCP) [9] . Fu et al. proposed a weighted variation
model for simultaneous reflection and illumination estimation (SRIE) that could preserve the
estimated reflectivity with high accuracy and suppress noise to a certain extent [10]. Image
enhancement methods based on machine learning have emerged in recent years [11]. The image
quality for different distortions lie noise, compression etc., is estimated [12].

Fig .1. Illumination Reflection model


F(x,y) = I(x,y) * R(x,y) [1]
F(x,y) represents the brightness of the image at particular pixel is the product of I(x,y) and R(x,y).
I(x,y) and R(x,y) represents illumination component of the incident light and transmission component
from object surface.
III. Methodology
Primarily, the RGB colour space is converted to HSV colour space. The V-plane in extracted and
information from brightness is improved and finally HSV plane is back converted into RGB plane.
The brightness plane of the image is enhanced using adaptive enhancement and histogram
equalization techniques followed with image fusion. The colour image transformation combined with
multi-scale decomposition is used for enhancing low illumination images.

Fig .2. Mechanism of enhancement


Unlike RGB colour space which has primary colour components, HSV is nearer to human perception
of colour. HSV colour space has three components: 1.Hue, 2. Saturation, 3. Value or brightness plane.
V = max (R, G, B) [2]
S = 1 − min (R, G, B) / V [3]
H:
= 60 * (G-B) / (V- min(R, G, B))
If V = R
= 120+60 * (B-R) / (V- min(R, G, B))
If V = G
= 240+60 * (R-G) / (V- min(R, G, B))
If V = B [4]

ISSN: 2005-4238 IJAST 10467


Copyright ⓒ 2020 SERSC
International Journal of Advanced Science and Technology
Vol. 29, No. 05, (2020), pp. 10466 - 10473

Let I be the image taken. Ih, Is, Iv represents three HSV colour planes.

Fig. 3. Hue, Saturation, Value planes of Box image.


The illumination component of the image is extracted from image by convolution of image with multi
scale Gaussian function.
Iv_g = ∑3𝑖=1 𝜃𝑖 (𝐼𝑣 ∗ 𝐺𝑖 (𝑥, 𝑦)) [5]
Iv_g represents estimated value of the illumination component and Iv represents brightness plane of the
image.
2 2 2
G(x, y) = λ𝑒 −(𝑥 +𝑦 )/𝜎 [6]
Where 𝜃𝑖 is weighting coefficient = 1/3
1
λ = Normalization factor =
√2𝜋𝜎 2
Gaussian function G(x, y) with three scale factors, σ = [15, 80, 250] is convoluted with image and
added to extract better illumination component even from unevenly illuminated images.
Image Enhancement:
1. Adaptive brightness enhancement:
The extracted reflection component is enhanced by modifying the distribution profile of the
illumination components of the image. The illumination values of over illuminated are decreased and
vice-versa. The brightness plane Iv is improved adaptively using the formula:
Iv (255+k)
I’v = [7]
Iv+k
Where 255 is the gray-level of 8-bit image, k is adjusting parameter. The magnitude of image
enhancement decreases with increase in value of k. To reduce over enhancement the formula is
further modified as,
Iv (255+k)
I’v = max(Iv,Iv_g)+k [8]
I’v is v-plane after enhancement.
k = α * mean (Is) where α = [0.1, 1]
The adaptive parameter is calculated by multiplying α with average value of the saturation plane. As
the ‘k’ increases the enhancement ability of the technique decreases.
2. Histogram Equalisation:
Histogram equalization is one of the basic image processing techniques used to improve the contrast
of the image. It is a contrast enhancement technique accomplished by adjusting image histogram. HE
assigns the intensity values of pixels in input image such that the output image has a uniform
distribution of intensities.
I’v = histequ (Iv) * max (Iv_g, Iv) [9]
Histequ represents histogram equalization. Image Fusion:
In image processing applications, the method used to extract key information from two or more sub
images is image fusion. The fused image is more informative than source images. It provides all the
necessary information needed. The image fusion is done as follows:
F = ∑𝑁𝑖=1 𝑤𝑖 𝑆𝑖 [10]
where F is fused image, w is the weighting coefficients, S is the source images, N is the total number
of images to be fused.
The sub images which are to be fused almost have similar information so the weighting coefficients
are obtained using PCA (Principal coefficient analysis). In PCA, the feature vectors of source images
are obtained and feature values are used to produce weighting coefficients.

ISSN: 2005-4238 IJAST 10468


Copyright ⓒ 2020 SERSC
International Journal of Advanced Science and Technology
Vol. 29, No. 05, (2020), pp. 10466 - 10473

Fig .4. Fusion methodology


Consider two source images S1, S2.

Fig.5. The two source images S1, S2 for two alpha values [0.01, 1] due to adaptive enhancement
technique.
1. Generate matrix S using both the source images.
S = [S1, S2]
2. Calculate the covariance matrix of S.
2 2
𝜎11 𝜎12
C= 2 2 [11]
𝜎21 𝜎22
𝜀1
3. Calculate the eigen values (λ1, λ2) and feature vector 𝜀 for covariance matrix C.
2
4. Find the largest eigen value.
5. Find the weighting coefficients w1 and w2 using feature values:
𝜀 𝜀
w1 = 𝜀 1 and w2 = 𝜀 2 [12, 13]
1+𝜀2 1+𝜀2
6. Finally calculate the fused image
F = w1S1 + w2S2 [14]
The fused V-plane is replaced with original V-plane and image is converted back into RGB colour
space.
IV. RESULTS AND DISCUSSIONS
In this paper, image enhancement for low illumination images is done using adaptive enhancement
and histogram equalization techniques. Six different low illuminated images are enhanced using these
two methods. The methods are tested using MATLAB 2016. The output enhanced images are clear
and bright. The adaptive enhancement is done using following algorithm1:
Algorithm1:

1. Read an image (I).


2. Convert RGB space to HSV space.
3. Extract the illumination component, Iv_g from Iv using Gaussian distribution [5].
4. Obtain two enhanced images Iv1 and Iv2 with adaptive mechanism using [8] taking α = 0.1 and
α=1.
5. Compute the covariance matrix (C) of Iv1 and Iv2 using [11].
6. Calculate the Eigen values and feature vector for C.
7. Find weighting factors w1 and w2 using [12, 13].
8. Apply the fusion formula to Iv1 and Iv2 using [14].
9. Pick up this fusion value plane, combine with Ih and Is and convert it into RGB space.
10. Display the enhanced image (J).

The six experimental images are ‘Box’, ‘Hill’, ‘Fort’, ‘Elephant’, ‘City’, ‘House’. The low
illuminated and enhanced images due to algorithm1 are shown below:

ISSN: 2005-4238 IJAST 10469


Copyright ⓒ 2020 SERSC
International Journal of Advanced Science and Technology
Vol. 29, No. 05, (2020), pp. 10466 - 10473

‘Box’

‘Hill’

‘Fort’

‘Elephant’

‘City’

‘House’
The enhancement for low illuminated images using histogram equalization is done by following
algorithm2:
Algorithm2:

1. Read an image (I).


2. Convert RGB space to HSV space.
3. Extract the illumination component, Iv_g from Iv using Gaussian distribution [5].
4. Obtain two enhanced images Iv1 and Iv2 with histogram equalization using [9].
5. Compute the covariance matrix (C) of Iv1 and Iv2 using [11].
6. Calculate the Eigen values and feature vector for C.
7. Find weighting factors w1 and w2 using [12, 13].
ISSN: 2005-4238 IJAST 10470
Copyright ⓒ 2020 SERSC
International Journal of Advanced Science and Technology
Vol. 29, No. 05, (2020), pp. 10466 - 10473

8. Apply the fusion formula to Iv1 and Iv2 using [14].


9. Pick up this fusion value plane, combine with Ih and Is and convert it into RGB space.
10. Display the enhanced image (J).

The same six images are enhanced using algorithm 2. The low illuminated and enhanced images due
to algorithm2 are shown below:

‘Box’

‘Hill’

‘Fort’

‘Elephant’

‘City’

‘House’
Quantitative analysis:
Entropy:
It is a statistical measure of randomness Entropy is used to characterize the texture of the input image.
It finds the amount of information available in the image. Let ‘A’ be the source set of symbols {a i}.
P(ai) represents the probability of occurrence of symbol {ai}.
E = ∑𝑁𝑖=1 𝑃(𝑎𝑖 ) log 2 𝑃(𝑎𝑖 )
N is the total number of symbols in source set ‘A’.
ISSN: 2005-4238 IJAST 10471
Copyright ⓒ 2020 SERSC
International Journal of Advanced Science and Technology
Vol. 29, No. 05, (2020), pp. 10466 - 10473

RMSE and PSNR:


The root mean-square error (RMSE) and the peak signal-to-noise ratio (PSNR) are used to compare
quality of enhanced image. The RMSE represent root mean square error between enhanced and the
original image, whereas PSNR represents a measure of the peak error. The lower the value of MSE,
lower the error.
MSE is calculated as follows:
1 𝑁 ̅
MSE= √𝑀𝑁 ∑𝑀
𝑖=1 ∑𝑗=1[𝑥 (𝑖, 𝑗) − 𝑥(𝑖, 𝑗)]
2

M, N represents the rows and columns of the image. MSE can be calculated for any dimensional
image. 𝑥̅ (i, j) and x(i, j) represents the two images to be compared .
PSNR is calculated as follows:
PSNR= 10log10 (peak2/MSE)
where ‘peak’ is either specified by the user or taken from the range of the image. ‘MSE’ is mean
square error between enhanced and input image. PSNR is calculated in dB.
The entropy, PSNR and RMSE is calculated from all the six images and shown below in tables - 1, 2.
Table-1: Entropy, PSNR, RMSE values for six enhanced images due to adaptive enhancement
technique (algorithm1).

Table-2: Entropy, PSNR, RMSE values for six enhanced images due to histogram equalization
technique (Algorithm 2).

V.CONCLUSIONS AND FUTURE SCOPE


In this paper, the low illuminated images are enhanced based on two enhancement techniques: 1.
Adaptive Enhancement, 2. Histogram Equalization. The correction of brightness is done in HSV
colour space. The V-plane is enhanced in both the techniques by modifying the distribution the
distribution profile of the reflection component. In adaptive enhancement technique the parameters
are adaptively adjusted to enhance the image brightness. In histogram equalisation method, the
contrast of the V-plane is enhanced. The enhancement methods are followed by fusion strategy to
further improve the information of image. The resulted enhanced images due to both the algorithms
are displayed. The processed images are more clear, and bright. The Quantitative analysis is also done
to the enhanced images. The entropy, PSNR, MSE of the images are calculated and tabulated. The
entropy of the images is improved, PSNR is increased and error is reduced. Compared to both the

ISSN: 2005-4238 IJAST 10472


Copyright ⓒ 2020 SERSC
International Journal of Advanced Science and Technology
Vol. 29, No. 05, (2020), pp. 10466 - 10473

methods adaptive enhancement technique provides better results in all aspects compared to histogram
equalization technique. This paper can be further extended to video processing.

References:
[1] Rafael C. Gonza Lez, Richard E. Woods ,”Digital Image Processing, 4th Edition”
[2] W. Wang, Z. Chen and X. Yuan et al., “Adaptive image enhancement method for correcting low-
illumination images “Information Sciences 496 (2019).

[3] T.Huynh-The, B. Le, S. Lee, et al.,Using weighted dynamic range for histogram equalization to
improve the image contrast, EURASIP J. Image Video Process. 1 (1) (2014) 44.
[4] S. Huang, F. Cheng, Y. Chiu, Efficient contrast enhancement using adaptive gamma correction
with weighting distribution, IEEE Trans. Image Process. 22 (3) (2013) 1032–1041.
[5] D. Jobson , Z. Rahman , G. Woodell , Properties and performance of a center/surround Retinex,
IEEE Trans. Image Process. 6 (3) (1997) 451–462 .
[6] E. Land, J. McCann , Lightness and retinex theory, J. Opt. Soc. Am. 61 (1) (1971) 1–11 .
[7] D. Jobson , Z. Rahman , G. Woodell , A multiscale Retinex for bridging the gap between color
images and the human observation of scenes, IEEE Trans. Image Process. 6 (7) (2002) 965–976.
[8] Z. Rahman , D. Jobson , G. Woodell , Retinex processing for automatic image enhancement, J.
Electron. Imaging 13 (1) (2004) 100–110.
[9] A. Petro, C. Sbert , J. Morel , Multiscale Retinex, Image Process on Line 4 (2014) 71–88.
[10] X. Fu , D. Zeng , Y. Huang , et al. , A weighted variational model for simultaneous reflectance
and illumination estimation, in: IEEE Conference on Computer Vision and Pattern Recognition,
2016, pp. 2782–2790 .
[11] M. Gharbi, J. Chen, J. Barron , et al. , Deep bilateral learning for real-time image enhancement,
ACM Trans. Graph. 36 (4) (2017) 118.
[12] Pinki, Rajesh Mehra, et al., “Estimation of the Image Quality under Different Distortions
”International Journal Of Engineering And Computer Science ISSN: 2319-7242 Volume 5
Issues 7 July 2016.
BIBLIOGRAPHY OF AUTHORS

T. V. Hyma Lakshmi obtained her


M.Tech.(ECE)., from JNTUCE,
Ananthapur and B.E. (ECE) from
S.R.K.R. Engineering college
Bhimavaram. Presently working
as Assistant Professor in S.R.K.R
.Engineering college Bhimavaram
and pursuing Ph.D. Program from

G.Satish completed his B.Tech


from DNR College of
Engineering and technology,
Bhimavaram. He is pursuing
his M.Tech in SRKR College,
Bhimavaram. His research
interests include digital signal
and image processing.

ISSN: 2005-4238 IJAST 10473


Copyright ⓒ 2020 SERSC

You might also like