Survey Paper on Various Image Fusion Techniques

Abstract- One of the wide research field in image processing is image fusion. Combining of all the information and data from all the images without any ruin of information and distortion is image fusion. To obtained an image having all the relevant features at focus is not easy, so get all the feature in single image is obtained by fusing the images with different techniques. This paper describe the brief survey of the different fusion techniques like Principal component analysis (PCA), Intensity Hue saturation (IHS) based method, Log Gabor wavelet transform, Pyramid algorithm, Shearlet Transform etc.

Keywords—Image Fusion, IHS, PCA, NSCT

INTRODUCTION

Image fusion is the process of amalgamate data of one or multiple images of the same scene without producing the detail which are not existent in the given images. The subsequent image will be more useful than any of the input images. Image fusion is a process of amalgamate relevant data of two or more images of the same scene. The aim of image fusion is to reducing the amount of data in network transmissions is to create new images that are more suitable for the purposes of human and machine perception, and for more image- processing. The input image can be multi view, multi-modal, multisensor or multi temporal. Images captured from the same image modality but different point of view is Multi view fusion[1]. Images captured at different sensors are Multimodal. Images captured at different instant of time is Multitemporal [2]. Image fusion can be define at different stages i.e Pixel, Feature and Decision. Pixel-based fusion is a low level of fusion and is depends on the pixel of the image. To improve the performance of the image pixel level fusion is performed in which information content with each pixel is obtained from set of pixel in source image [3]. Feature level is a middle level of fusion requires an extraction of objects recognized in the various data sources. It requires the extraction of important features which are depending on their environment such as pixel intensities, textures, edges etc. These similar features from input images are fused[4]. Decision-level fusion consists of merging information at a higher level of abstraction, combines the results from several algorithms to yield a final fused decision. Input images are procesed individually for extraction of information[5]. The information we obtained is then combined applying decision rules to increase ,furth common interpretation.

LITRATURE SURVEY

Image fusion is the very wide and important topic of image processing. Various method have been used to reduced the blurring effect and enhance the quality feature of the image. The work on image fusion is started in mid of eighties.the litraure survey is as follows

Burt use the laplacian Pyramid to compute the image fusion. Pyramid algorithm is very widely used in image fusion[6].

Adlelson use the depth of focus fusion method for the fusion of two image[7].

Cheng-I Chen compute the combination of IHS and Log Gabor wavelet transform for the fusion of MRI and PET image and reduced the color distortion in the fused image[8].

Jia Du,Weisheng Li and Bin Xiao uses an novel approach of functional image fusion by the information of interest in local laplacian filtering. In this approach Based on number of levels input anatomical medical image and functional medical images are decomposed into multi scale image which gives the better performance as compared to the other fusion techniques [9].

For the fusion of Medical Image fusion and Denoising with Alternating Sequential Filter Wenda Zhao and Huchuan Lu computed the Adaptive fractional order total variation method which supperssing the noise while avoiding the staircase effect of the total variation[10].

Yong Yang,Yue Que, shuying Huang and Pan Lin presented combination of Non subsampled counterlet transform andType-2 fuzzy logic to preserve the more information of the fused image. It also improve the quality of the fused image [11].

Vikrant Bhateja, Himanshi Patel, Abhinav Krishn, Akanksha sahu, Aime Lay-Ekuakille implemented -Stationary wavelet transform and Non subsampled counterlet transform for increasing the frequency , time localization with shift variance, minimization of redundancy, better restoration and improved the contrast of the fused image [12].

Sudeb Das and Malay kumar kundu propose a new method. In this paper strengths of neurons in the RPCCN are adaptively computed based on the fuzzy characteristics of the image,method can preserve more useful information in the fused image with high spatial resolution and less difference to the source image[13].

Lei Wang, Bin Li and Lianfang Tian have proposed the multimodal medical image fusion. The author have used the 3D Shearlet transform for the fusion of MRI brain images of normal brain with the MRI brain images with noise in this shearlet transform and wavelet transform are used in the same circle. In this circle can be decomposed in to more high pass subbands in each levels and than the only vertical, horizontal and diagonal subband of the wavelet transform. So the more features information and directional sensitivity in different levels can be captured by the shearlet transform. The 3D shearlet transform consist of two levels 3-D Laplacian pyramid filter for the multiscale partition and pseudo-spherical Fourier transform for the directional localization. 3D shearlet transform provide the better image representation of fused image with high quality [14].

Gaurav Bhatnagar, Q.M.Jonathan wu and Zheng Lui have worked on the Non –Subsampled contour transform. In this paper the multimodal medical image (i.e MRI ,PET) have been fused. MRI is a panchromatic image and the PET is a multispectral image. In the NSCT domain the source images are first converted in to high frequency and low frequency sub bands. After that low frequency and high frequency fusion rule is applied in each bands. Than we get the low frequency and high frequency fused image . afterthat they apply the inverse NSCT to obtained the fused image. The fused image Enhance the details of fused images and improve the visual effect with much less information distortion [15].

He, D et al. explained that the challenge in image fusion is to fuse two types of images by forming new images integrating both the spectral aspects of the low resolution images and the spatial aspects of the high resolution image[16].

Rui shen, Irene Cheng and Anup basu have proposed the Cross scale fusion rule for the volumetaric medical image fusion.This paper demonstrate the effectiveness and versatility of the fusion rule. The resultant image is of the high quality [17].

In this paper Vince D.Calhoun and Tulay Adali have explain the Independent Component Analysis and multivariate Data analysis for the fusion of medical images[18].

Lavanya, A. et al. proposed a image fusion method based on wavelet combined IHS and PCA transformations for remotely sensed lunar images in order to extract features accurately[19]

A.Soma Sekhar et al. has proposed a Novel multi-resolution fusion algorithm for medical diagnosis using integrated PCA and wavelet transforms .A multi-resolution based fusion is obtained by combining the aspects of region and pixel-based fusion [20].

V.P.S. Naidu has perform the new method Multi-resolution singular value decomposition for the image fusion. This method is used for smoothing the fused image [21]

OBJECTIVE OF VARIOUS IMAGE FUSION ALGORITHM

The objective of various image fusion algorithm are as follows

• To extract the important information from the fused image
• Algorithms should be reliable and robust imperfections such as mis-registration.
• Image fusion algorithim should be cost effective to acquisition of information.