GRD Journals- Global Research and Development Journal for Engineering | Volume 4 | Issue 9 | August 2019

ISSN: 2455-5703

Survey on Image Fusion Techniques

Nav Jyoti Prakash Pandey Mr. Apoorv Mishra
Research Scholar Assistant Professor
Department of Computer Science and Engineering Department of Computer Science and Engineering
Naraina College of Engineering and Technology Naraina College of Engineering and Technology

Abstract
Image fusion is process of combining two different images of same scene which are multi focused by nature. It has the major
application in the field of visual sensor network for efficient surveillance monitoring. The thesis presents a novel hybrid image
fusion technique that has the capabilities to be used in the real time environment such as central computer of visual sensor network
for efficient surveillance purpose. A high resolution panchromatic image gives geometric details of an image because of the
presence of natural as well as manmade objects in the scene and a low resolution multispectral image gives the color information
of the source image. The aim of multisensor image fusion is to represent the visual information from multiple images having
different geometric representations into a single resultant image without any information loss. The advantages of image fusion
include image sharpening, feature enhancement, improved classification, and creation of stereo data sets. Multisensor image fusion
provides the benefits in terms of range of operation, spatial and temporal characteristics, system performance, reduced ambiguity
and improved reliability. Based on the processing levels, image fusion techniques can be divided into different categories. These
are pixel level, feature level and symbol level/decision level. Pixel level method is the simplest and widely used method. This
method processes pixels in the source image and retains most of the original image information. Compared to other two methods
pixel level image fusion gives more accurate results. Feature level method processes the characteristics of the source image. This
method can be used with the decision level method to fuse images effectively. Because of the reduced data size, it is easier to
compress and transmit the data. The top level of image fusion is decision making level. It uses the data information extracted from
the pixel level fusion or the feature level fusion to make optimal decision to achieve a specific objective. Moreover, it reduces the
redundancy and uncertain information.
Keywords- Image Fusion, Comparative Analysis, Quantitative Analysis, Computational Analysis etc

I. INTRODUCTION
Generally two sorts of vision are described. They are human vision and PC vision. Human vision is refined system that resources
and follows up on visual updates. It has progressed for a long time, essentially for obstruction or survival. Basic PC vision structure
requires a camera, a camera interface and a PC.
Image combination is the way toward consolidating at least two or more than two pictures into the single picture to
upgrade the data content. Picture combination systems are significantly utilized in applications remote detecting, military,
stargazing and therapeutic imaging. Picture blend systems are noteworthy as it improves the introduction of article affirmation
structures by fusing various wellsprings of satellite, airborne and ground based imaging structures with other related instructive
files.

II. LITERATURE REVIEW

A huge amount of research and work has been done on picture blend methodologies since mid-nineteen eighties. The most direct
strategy for consolidating pictures is by taking the diminish scale ordinary of the pixels of source pictures. This direct technique
gives extraordinary results to the detriment of lessened intricacy level. These techniques of merging pictures can be associated
with different educational accumulations depending on their spatial and transient characteristics. Spatial zone frameworks and
Frequency region techniques are used for joining the photos. Spatial region methodology procedure picture pixels to achieve the
perfect result while repeat space frameworks first trade the image into repeat zone by applying Fourier change and after that procure
the resultant picture by performing inverse Fourier change. These procedures are taken a gander at using execution estimation
qualities, for instance, Peak signal to Noise ratio (PSNR), Mean square error, and entropy.
1) IHS (Intensity-Hue-Saturation) Transform technique.
2) Principal Component Analysis (PCA) technique.
3) Pyramid technique.
4) High pass filtering technique.
5) Wavelet Transform technique.
6) Artificial Neural Networks technique
7) Pyramid Technique

All rights reserved by www.grdjournals.com 1

 Survey on Image Fusion Techniques
(GRDJE/ Volume 4 / Issue 9 / 001)

A. Intensity Hue Saturation Transform

Power, Hue and Saturation are the three properties of shading that give controlled visual depiction of an image [9]. IHS change
procedure is the most prepared strategy for picture mix. In the IHS space, shade and inundation ought to be circumspectly controlled
in light of the fact that it contains an enormous bit of the apparition information. For the blend of high objectives PAN picture and
multispectral pictures, the detail information of high spatial objectives is added to the absurd information.

B. Principal Component Analysis (PCA) Technique

Picture pyramids can be portrayed as a model for the binocular mix for human visual system. By molding the pyramid structure a
remarkable picture is addressed in different measurements. A composite picture is encircled by applying a model explicit
methodology of picture mix. Immediately, the pyramid rot is performed on each source picture. All of these photos are joined to
outline a composite picture and a short time later talk pyramid change is associated with get the resultant picture [4].

C. High Pass Filtering Technique

The high objectives multispectral pictures are gotten from high pass isolating. The high repeat information from the high objectives
panchromatic picture is added to the low objectives multispectral picture to get the resultant picture. It is performed either by
isolating the High Resolution Panchromatic Image with a high pass channel or by taking the one of a kind HRPI and subtracting
LRPI from it [8]. The apparition information contained in the low repeat information of the HRMI is shielded by this strategy.

D. Principal Component Analysis Technique

Notwithstanding resembling IHS change, the advantage of PCA system over IHS methodology is that a self-decisive number of
gatherings can be used. This is a champion among the most unmistakable systems for picture mix. Uncorrelated Principal fragments
are molded from the low objectives multispectral pictures [7]. The principal fundamental part (P1) has the information that is
essential to all gatherings used. It contains high distinction to such a degree, that it gives more information about panchromatic
picture. A high objectives PAN portion is stretched out to have a similar change as P1 and replaces P1. By then a retrogressive
PCA change is used to get the high objectives multispectral picture.

E. Wavelet Transform Technique

Wavelet change is considered as an alternative as opposed to the brief timeframe Fourier changes. It is advantageous over Fourier
change in that it gives needed objectives in time zone similarly as in repeat region however Fourier change gives adecent objectives
in just repeat region. In Fourier change, the sign is deteriorated into sine surges of different frequencies while the wavelet change
rots the sign into scaled and moved kinds of the mother wavelet or limit. In the image blend using wavelet change, the data pictures
are decayed into deduced what's increasingly, illuminating coefficients using DWT at some specific dimension [6]. A blend rule
is associated with join these two coefficients and the resultant picture is gotten by taking the switch wavelet change.

F. Discrete Cosine Transform-DCT Technique

Discrete cosine Transform has discovered significance for the compacted pictures as MPEG, JVT and so on. By taking discrete
cosine change, the spatial space picture is changed over into the recurrence space picture [3]. Normal brightening is spoken to by
the DC esteem and the Air conditioning esteems are the coefficients of high recurrence. The RGB picture is separated into the
squares of with the size of 6*6 pixels. The picture is then assembled by the frameworks of red, green and blue and changed to the
dim scale picture.

G. Artificial Neural Networks- ANN Technique

ANN have found their essentialness in precedent affirmation. In this a nonlinear response limit is used [9]. It uses Pulse Coupled
Neural Network (PCNN) which includes an analysis orchestrate. This framework is disengaged into three segments to be explicit
the responsive field, the change field and the beat generator. Each neuron identifies with the pixel of the data picture. The organizing
pixel's capacity is used as an outside commitment to the PCNN. This procedure is beneficial to the extent quality against uproar,
self-governance of geometric assortments and capacity of traverse minor power assortments in information plans. PCNN has
common centrality and used in helpful imaging as this system is achievable and gives persistent structure execution [10].

III. COMPARATIVE ANALYSIS

The experimental results are evaluated on the standard Lytro dataset available at open access database. This dataset is of multi-
focus image. The testing is performed on the grayscale images to easily analyse the testing results. The image size resolution is set
at the 256×256 pixels. The experimental results of the proposed method are compared with some of the major and most prevalent
available techniques like Averaging, DWT (1-level), DWT (3-level), SWT and PCA. The final fusion results are compared using
qualitative as well as quantitative analysis. Also the proposed method is also compared on the basis of the execution time. All the
compared methods and proposed method are tested on the system of same configuration. This is required for the evaluation of the
computational efficiency.

All rights reserved by www.grdjournals.com 2

 Survey on Image Fusion Techniques
(GRDJE/ Volume 4 / Issue 9 / 001)

A. Quantitative Analysis
The experimental testing is performed on overall all the multi-focus images of the database. But here in the thesis, the results are
presented on the limited dataset. The Table 1 and Figure 1 comparatively analyses the result of proposed method with other standard
methods on the basis of quantitative results. On comparing the results in the Table 1, it is observed that the bold values of the
proposed method shows the best results in terms of quantitative analysis. The overall image quality of the proposed method results
is also far better than all other methods.
Table 1: Analysed using graphically for more better and easy understanding and analysis
Methods SD CC API AG He MI QAB/F
Averaging 50.115 0.9131 96.012 3.686 4.265 3.125 0.798
DWT (1-level) 50.302 0.9254 98.066 4.598 4.258 5.125 0.888
DWT (3-level) 50.412 0.9472 97.118 3.265 4.569 4.989 0.799
SWT 51.052 0.9451 98.012 4.235 4.111 4.119 0.826
PCA 51.312 0.9543 99.091 4.598 5.020 4.989 0.581
Proposed method 52.521 0.9699 100.871 5.551 5.111 5.811 0.955
Table 1 Average results (without reference) on gray- scale dataset. The best results are shown in bold.

B. Computational Analysis
Table 2: compares the execution time of proposed method with other standard methods
Methods Time (in seconds)
Averaging 1.1155
DWT (1-level) 3.1302
DWT (3-level) 7.0023
SWT 4.0572
PCA 5.3712
Proposed method 3.2365
Table 2 Average results (without reference) on grayscale dataset. The best results are shown in bold.

Fig. 1: Comparative computational analysis

IV. CONCLUSION AND FUTURE SCOPE

Image fusion is process of combining two different images of same scene which are multifocused by nature. It has the major
application in the field of visual sensor network for efficient surveillance monitoring. The thesis presents a novel hybrid image
fusion technique that has the capabilities to be used in the real time environment such as central computer of visual sensor network
for efficient surveillance purpose.
The analysis on various dataset of different sizes from different databases. It is found that results of different methods
shows the best fusion results on comparing with some of the most prevailing standard methods. The experimental results are
performed on the basis of qualitative and quantitative analysis. The comparative computational time is also evaluated for better
judgment of the effectiveness of proposed method.
This paper analysis the various techniques of image fusion such as DWT SWT PCA LDA etc. On analyzing it is observed
that the methods related tp wavelet transform are better compare to other methods due to high level of robustness and adaptive
nature of wavelet transform.
The wavelet transform are prevalent now a days because of there is lot of scope to perform and incorporate other
technologies in it. This paper also analyses the results of different methods using qualitative and quantitative analysis.

All rights reserved by www.grdjournals.com 3

 Survey on Image Fusion Techniques
(GRDJE/ Volume 4 / Issue 9 / 001)

REFERENCES
[1] Liu, Zhaodong, Yi Chai, Hongpeng Yin, Jiayi Zhou, and Zhiqin Zhu. "A novel multi-focus image fusion approach based on image decomposition."
Information Fusion 35 (2017): 102-116.
[2] M. Nejati, S. Samavi, and S. Shirani, "Multi-focus Image Fusion Using Dictionary-Based Sparse Representation", Information Fusion, vol. 25, Sept. 2015,
pp. 72-84.
[3] Multi focus image dataset, Available at: http://mansournejati.ece.iut.ac.ir/content/lytro-multi-focus-dataset.
[4] Y. Liu, S. Liu, Z. Wang, A general framework for image fusion based on multi-scale transform and sparse representation, Inf. Fusion 24 (2015) 147–164.
[5] B.K.S. Kumar, Image fusion based on pixel significance using cross bilateral filter, Signal Image Video Process. 9 (5) (2015) 1193–1204.
[6] S. Paul, I.S. Sevcenco, P. Agathoklis, Multi-exposure and multi-focus image fusion in gradient domain, J. Circuit Sys. Comp. (2016) 1650123.
[7] K. Zhan, J. Teng, Q. Li, J. Shi, A novel explicit multi-focus image fusion method, J. Inf. Hiding Multimedia Signal Process. 6 (3) (2015) 600–612.
[8] Farid, M.S., Mahmood, A. and Al-Maadeed, S.A., 2019. Multi-focus image fusion using Content Adaptive Blurring. Information Fusion, 45, pp.96-112.
[9] F. Samadzadegan, F. DadrasJavan, “Evaluating the Sensitivity of Image Fusion Quality Metrics to Image Degradation in Satellite Imagery”, Journal of the
Indian Society of Remote Sensing • December 2011, DOI: 10.1007/s12524-011-0117-z
[10] Basic Intensity Quantification with ImageJ, Available at
https://www.unige.ch/medecine/bioimaging/files/1914/1208/6000/Quantification.pdf
[11] Multi focus image dataset, Available at: http://dsp.etfbl.net/mif/
[12] V.P.S. Naidu, “Image Fusion technique using Multi-resolution singular Value decomposition”, Defence Science Journal, Vol. 61, No. 5, September 2011,
pp. 479-484.

All rights reserved by www.grdjournals.com 4