Pixel Level Image Fusion A Review on Various Techniques

Transcription

1 3 rd World Conference on Applied Sciences, Engineering & Technology September 2014, Kathmandu, Nepal Pixel Level Image Fusion A Review on Various Techniques JAGALINGAM P., ARKAL VITTAL HEGDE Department of Applied Mechanics & Hydraulics, National Institute of Technology Surathkal, Karnataka, India lingam.jai10@gmail.com, arkalvittal@gmail.com Abstract: Image fusion integrates the spatial information of high-resolution panchromatic (PAN) image and the color information of a low resolution multispectral (MS) image to produce high resolution MS image. The technique generates a single new enhanced image that provides richer information than each of the individual images. Image fusion can be performed at different levels of information representation, namely: pixel level, feature level and decision level. Almost all image fusion algorithms developed to-date, work only at pixel level. One of the keys to image fusion algorithms is how effectively and completely to represent the source images. The present paper reviews some of the successful pixel level image fusion algorithms such as principal component analysis, intensity hue saturation, brovey transform and multi-scale transform etc., and discusses their strengths and weaknesses. Keywords: Pixel level image fusion, Principal component analysis, Intensity hue saturation, Brovey transform, Multiscale transform. Introduction: In the mid 1980s, image fusion received significant attention from researchers in remote sensing and image processing. Image fusion also called Pan sharpening, is a technique used to integrate the geometric details of a high resolution panchromatic (PAN) image and the color information of a lowresolution multispectral (MS) image to produce a high-resolution MS image. This technique is particularly important for large-scale applications. Why is image fusion important? Most earth resource satellites, such as SPOT, IRS, Landsat 7, IKONOS, QuickBird and OrbView, plus some modern airborne sensors, such as Leica ADS40, provide both PAN Images at a higher spatial resolution and MS images at a lower spatial resolution. An effective image fusion technique can virtually extend the application potential of such remotely sensed images, as many remote sensing applications require both high-spatial and high-spectral resolutions, especially for GIS based applications. Why don t most satellites collect high-resolution MS images directly, to meet this requirement for high-spatial and high-spectral resolutions? There are two major technical limitations involved: (1) the incoming radiation energy to the sensor, and (2) the data volume collected by the sensor. In general, a PAN image covers a broader wavelength range, while a MS band covers a narrower spectral range. To receive the same amount of incoming energy, the size of a PAN detector can be smaller than that of a MS detector. Therefore, on the same satellite or airplane platform, the resolution of the PAN sensor can be higher than that of the MS sensor. In addition, the data volume of a highresolution MS image is significantly greater than that of a bundled high-resolution PAN image and lowresolution MS image. This bundled solution can mitigate the problems of limited on-board storage capacity and limited data transmission rates from platform to ground. Considering these limitations, it is clear that the most effective solution for providing high-spatial-resolution and high-spectral-resolution remote sensing images is to develop effective image fusion techniques. There has been growing interest in the use of multiple sensors to increase the capabilities of intelligent machines and systems. Multi-sensor fusion refers to the synergistic combination of different sources of sensors information into a single representational format. Multi-sensor fusion can occur at the signal level, image level, feature level and symbol level of representation. Signal-level fusion refers to the direct combination of several signals in order to provide a signal that has the same general format as the source signals. [1][2]Image-level fusion (also called pixel-level fusion) generates a fused image in which each pixel is determined from a set of pixels in each source image. Feature level fusion first employs feature extraction on the source data so that features from each source can be jointly employed for some purposes. Symbol-level fusion allows the information from multiple sensors to be effectively combined at the highest level of abstraction. A common type of symbol-level fusion is decision fusion. Most common sensors provide data that can be fused at one or more of these levels. The different levels of multi sensor fusion can be used to provide information to a system that can be used for a variety of purposes. Table1 shows comparison of image fusion performance at various levels. To produce hybrid (fused) images with good quality some aspects should be considered during the fusion process and they are: 1) The PAN and MS images should be acquired at nearby dates. Several changes may occur during the interval of acquisition time. Variations in the vegetation depending on the season of the year, different lighting conditions, construction of buildings, or changes caused by natural catastrophes (e.g. earthquakes, floods and volcanic eruptions). WCSET BASHA RESEARCH CENTRE. All rights reserved.

2 JAGALINGAM P., ARKAL VITTAL HEGDE 2)The spectral range of PAN image should cover the spectral range of all multispectral bands involved in the fusion process to preserve the image color. This condition can avoid the color distortion in the fused image. 3) The spectral band of the high resolution image should be as similar as possible to that of the replaced low resolution component in the fusion process. Currently, pixel-level image fusion systems are used extensively in the fields of military, medical imaging and remote sensing. Table 1: comparison of image fusion performance at various levels Pixel level Feature Decision Level level level Feature The amount maximum medium minimum of information Information minimum medium maximum loss Dependence maximum medium minimum of the sensor Immunity the worst medium the best Detection Performance the best medium the worst Pixel level image fusion: Pixel level fusion can be used to increase the information content associated with each pixel in an image formed through a combination of multiple images, e.g., the fusion of a range image with a twodimensional intensity image adds depth information to each pixel in the intensity image that can be useful in the subsequent processing of the image. Different images to be fused can come from a single imaging sensor or a group of sensors. The fused image can be created either through the pixel-by-pixel fusion or through the fusion of associated local neighbourhoods of pixels in each of the images. The improvement in quality associated with pixel-level fusion can most easily be assessed through the improvements noted in the performance of image processing tasks such as (segmentation, feature extraction, and restoration) when the fused image is used to compare the use of the individual images. The fusion of multisensory data at the pixel level can serve to increase the useful information content of an image so that more reliable segmentation can take place and more discriminating features can be extracted for further processing. Pixellevel fusion can take place at various levels of representation: the fusion of the raw signals from multiple sensors prior to their association with a specific pixel, the fusion of corresponding pixels in multiple registered images to from a composite or fused image, and the use of corresponding pixels or local groups of pixels in multiple registered images for segmentation and pixel-level feature extraction. Fusion at the pixel level is useful in terms of total system processing requirements because fusion is made of the multisensory data prior to processingintensive functions like feature matching, and can serve to increase overall performance in tasks like object recognition because the presence of certain substructures like edges in an image from one sensor usually indicates their presence in an image from another sufficiently similar sensor. In order for pixellevel fusion to be feasible, the data provided by each sensor must be able to be registered at the pixel level and in most cases, must be sufficiently similar in terms of its resolution and information content. Although it is possible to use many of the general multisensory fusion methods for pixel level image fusion. The paper aims at discussing some of the successful pixel level image fusion algorithms. Intensity hue saturation: The colour system with red, green and blue channels (RGB) is usually used by computer monitors to display a colour image. Another color system widely used to describe a colour is the system of intensity, hue and saturation (IHS) [3][4][5]. The intensity represents the total amount of the light in a colour, the hue is the property of a colour determined by its wavelength, and the saturation is the purity of the colour. The IHS transform is always applied to an RGB composite. This implies that the fusion will be applied to groups of three bands of the MS image. As a result of this transformation, we obtain the new intensity, hue, and saturation components. The I component can be deemed as an image without color information. Because the I component resembles the PAN image, The PAN image then replaces the intensity image. Before doing this, in order to minimize the modification of the spectral information of the fused MS image with respect to the original MS image, the histogram of the PAN image is matched with that of the intensity image. Applying the inverse transform, we obtain the fused RGB image color model as shown in (Figure 1) with the spatial detail of the PAN image incorporated into it. The IHS technique is fairly easy to understand and implement. Moreover, it requires very little computation time compared to the other techniques. Although, the IHS method has been widely used, the method cannot decompose an image into different frequencies in frequency space such as higher or lower frequency. Hence the IHS method cannot be used to enhance certain image characteristics. However, it severely distorts the spectral values of the original colour of the MS image. Thus IHS technique is good only for visual analysis, and not for machine classification based on the spectral signatures of the original MS image. Moreover, it is also limited to three bands at a time. 3.1 Procedure to perform IHS Algorithm: (1) Perform image registration (IR) to PAN and MS, and resample MS. (2) Convert MS from RGB space into IHS space.

3 Pixel Level Image Fusion A Review on various techniques (3) Match the histogram of PAN to the histogram of the I component. (4) Replace the I component with PAN. (5) Convert the fused MS back to RGB space. Principal component analysis: The limitation of IHS leads to the development of Principal Component analysis. The fusion method based on PCA is very simple. [6][7] PCA is a general statistical technique that transforms multivariate data with correlated variables into one with uncorrelated variables. These new variables are obtained as linear combinations of the original variables. PCA has been widely used in image encoding, image data compression, image enhancement and image fusion. In the fusion process, PCA method generates uncorrelated images (PC1, PC2, PCn, where n is the number of input multispectral bands). In general, the PC1 collects the spatial information which is common to all the bands, while PC2, PC3 collect the spectral information that is specific to each band. The first principal component (PC1) is replaced with the panchromatic band, which has higher spatial resolution than the multispectral images. Later, the inverse PCA transformation is applied to obtain the image in the RGB color model as shown in (Figure 2). Therefore, PCA based fusion is very suitable for merging the MS and PAN images. Compared to the IHS fusion, the PCA fusion has the advantage that it does not have the three band limitation and can be applied to any number of bands at a time. However, this technique also introduces spectral distortion in the fused image like the IHS method. 4.1 Procedure to Perform PCA Algorithm: (1) Perform IR to PAN and MS, and resample MS. (2) Convert the MS bands into PC1, PC2, PC3, by PCA transform. (3) Match the histogram of PAN to the histogram of PC1. (4) Replace PC1 with PAN. (5) Convert PAN, PC2, PC3, back by reverse PCA. Figure 1: Standard IHS fusion scheme Figure 2: Standard PCA fusion scheme 5. Brovey Transforms (BT): The BT promoted by an American scientist, Brovey, is also called the colour normalization transform, is based on the chromaticity transform and the concept of intensity modulation. [9][19] It is a simple method to merge data from different sensors, which can preserve the relative spectral contributions of each pixel but replace its overall brightness with the high spatial resolution image. As applied to three MS bands each of the three spectral components (as RGB components) is multiplied by the ratio of a high resolution co- registered image to the intensity component I of the MS data. It is operated by the formula. IMG lowi IMG i = IMG highi=1, 2, 3 I Where IMG lowi, i = 1, 2, 3 denote three selected MS band image, IMG high for high-resolution image, and IMG i for fusion image corresponding to IMG lowi, i=1, 2, 3 and the intensity component refers to I= ( IMG low1 + IMG low2 + IMG low3 ) 3. It is evident that BT is indeed simple fusion methods requiring only arithmetical operations without any statistical analysis of filter design. The Brovey transform was developed to provide contrast in features such shadows, water and high reflectance areas. Interms of their efficiency and implementation, either of them can achieve the goal of fast fusion. However, color distortion problems are often produced in the fused images. Other arithmetic methods such as synthetic variable ratio (SVR) and ratio enhanced technique are similar but involve more sophisticated calculations for the MS sum for better fusion quality. 6. Multiscale Transform: The standard image fusion techniques, such as IHS, PCA and Brovey transform method operate under spatial domain. However, the spatial domain fusions produce spectral distortion. This is particularly crucial in optical remote sensing if the images to fuse were not acquired at the same time. Therefore, compared with the ideal output of the fusion, these methods often produce poor result. Over the past decade, new approaches or improvements on the existing approaches are regularly being proposed to overcome the problems in the standard techniques. As multiscale

4 JAGALINGAM P., ARKAL VITTAL HEGDE analysis has become one of the most promising methods in image processing, the multiscale transform such as pyramid methods, wavelet transform has become a very useful tool for image fusion. The techniques outperform the standard fusion techniques in spatial and spectral quality, especially in minimizing color distortion. The basic idea of multiscale transform is to perform a multi-resolution decompositions on each source image, then integrate all these decompositions to produce a composite representation. Multiscale transform such as pyramid methods, wavelet transform are discussed below. 6.1 Pyramid Method: One effective and pellucid structure used to describe image with multi-resolution is the image pyramid proposed by Burt and Adelson (1983). The basic principle of this method is to decompose the original image into pieces of sub-images with different spatial resolutions through some mathematical operations. A pyramid structure is an efficient organization methodology for implementing multiscale representation and computation. A pyramid structure can be described as a collection of images at different scales which together represent the original image. One of the most frequently used versions of the pyramid transform is the gaussian pyramid and laplacian pyramid. Other typical pyramids include morphological method, contrast pyramid, ratio of low pass pyramid method (RoLP), gradient method and filter-subtract-decimate method Laplacian Pyramid: The Laplacian pyramid [10] is derived from the gaussian pyramid, which is a multi-scale representation obtained through a recursive low-pass filtering and decimation. The laplacian pyramid decomposition is divided into two steps: the first is gaussian pyramid decomposition; the second is from gaussian pyramid to laplacian pyramid. Each level of the Laplacian pyramid is recursively constructed from its lower level by the following four basic procedures: Blurring (low-pass filtering), subsampling (reduce size), interpolation (expand in size) and differencing (to subtract two images pixel-by-pixel. In fact, a blurred and subsampled image is produced by the first two procedures at each decomposition level, and these partial results, taken from different decomposition levels, can be used to construct a pyramid known as the gaussian pyramid [19]. In both the laplacian and gaussian pyramids, the lowest level of the pyramid is constructed from the original image. In computing the laplacian and gaussian pyramids, the blurring is achieved using a convolution mask. Convolution is the basic operation of most image analysis system. In a multiresolution system one wishes to perform convolutions with kernels of many sizes, ranging from very small to very large. Let G k be the kth level of the gaussian pyramid for the image I. Then G o I and for k > 0, G k = [ G K-1 ] 2 Where [ 2] denotes down sampling of the signal by 2, which means to keep one sample out of two. This downsampling operation is performed in both horizontal and vertical directions. The kth level of the Laplacian pyramid is defind as the weighted difference between successive levels of the Gaussian pyramid. L k = G k - 4 [G k+1 ] 2 where [ 2] denotes upsampling, which means to insert a zero between every two samples in the signal. This upsampling operation is also performed in both horizontal and vertical directions. Here, convolution by has the effect of interpolating the inserted zero samples. An image can be reconstructed by the reverse procedure. Let G be the recovered Gaussian pyramid. Reconstruction requires all levels of the laplacian pyramid, as well as the top level of the Gaussian pyramid G N. Thus the procedure is to set G N =G N and for k < N, G k = L k +4 [G k+1 ] 2 which can be used to compute G 0, the reconstructed version of the original image G Fusion using morphological pyramid method: A morphological pyramid is obtained by applying morphological filters to the Gaussian pyramid at each level and taking the difference between 2 neighboring levels [11]. A morphological filer is usually for noise removal and image smoothing. It is similar to the effect of a low-pass filter, but it does not alter shapes and locations of objects in the image. The morphological pyramid fusion is therefore the same as the fusion using Laplacian pyramid method except replacing the Laplacian pyramid with the morphological pyramid Fusion using contrast pyramid: In the RoLP pyramid method, simply replace the division with the following contrast formula to obtain the contrast pyramids [13]. Con k = G k Expand (G k+1 ) Expand (G k+1) Where Con k represents the contrast between two successive levels G k and G k+1 in the Gaussian pyramids, and operation Expand consists of a simple upsampling followed by low-pass filtering Fusion using ratio of low pass pyramid method (RoLP): In the above laplacian pyramid method, simply replace the difference (i.e.., the Laplacian pyramid) with a division operation to obtain RoLP pyramids [14] Fusion using gradient pyramid method: A gradient pyramid is obtained by applying a set of 4 directional gradient filters (horizontal, vertical and 2 diagonal) to the gaussian pyramid at each level [16][20]. At each level, these 4 directional gradient pyramids are combined together to obtain a combined

5 Pixel Level Image Fusion A Review on various techniques gradient pyramid that is similar to a Laplacian pyramid. The gradient pyramid fusion is therefore the same as the fusion using laplacian pyramid method except replacing the laplacian pyramid with the combined gradient pyramid Fusion using filter-subtract-decimate (FSD) pyramid method: The FSD pyramid fusion method is conceptually identical to the Laplacian pyramid fusion method. The only difference is in the step of obtaining the difference images in creating the pyramid [18]. In Laplacian pyramid, the difference image L k at level k is obtained by subtracting an image upsampled and then low-pass filtered from level k+1 from the gaussian image G k at level k, while in FSD pyramid, this difference image is obtained directly from the Gaussian image G k at level k subtracted by the lowpass filtered image of G k. It is therefore obvious that FSD pyramid fusion method is computationally more efficient than the laplacian pyramid method by skipping an upsampling step. Figure 3: One-level two-dimensional discrete wavelet transform. a) Filterbank representation. b) Image representation. 6.2 Wavelet Transform: Wavelet transforms provide a framework in which a signal is decomposed, with each level corresponding to a coarser resolution, or lower frequency band [23]. There are two main groups of transforms, continuous and discrete. Continuous wavelet transform is simple to describe mathematically, both the signal and the wavelet function must have closed forms, making it difficult or impractical to apply. The discerete wavelet is used instead. The term discrete wavelet transform (DWT) is a general term, encompassing several different methods. It must be noted that the signal itself is continuous; discrete refers to discrete sets of dilation and translation factors and discrete sampling of the signal. The process of applying the DWT can be presented as a bank of filters. At each level of decomposition, the signal is split into high frequency and low frequency components; the low frequency components can be further decomposed until the desired resolution is reached. When multiple levels of decomposition are applied, the process is referred to as multiresloution decomposition. In practice when wavelet decomposition is used for image fusion, one level of decomposition can be sufficient, but this depends on the ratio of the spatial resolutions of the images being fused. Wavelet transformations, most fall into one of the following three categories: decimated, undecimated, and nonseparated Decimated: The conventional DWT can be applied using either a decimated or an undecimated algorithm. In the decimated algorithm, the signal is downsampled after each level of transformation. In the case of a twodimensional image, down-sampling is performed by keeping one out of every two rows and columns, making the transformed image one quarter of the original size and half the original resolution[28]. The decimated algorithm can therefore be represented visually as a pyramid, where the spatial resolution becomes coarser as the image becomes smaller. The wavelet and scaling filters are one-dimensional, necessitating a two-stage process for each level in the multiresolution analysis: the filtering and downsampling are first applied to the rows of the image and then to its columns. This produces four images at the lower resolution, one approximation image and three wavelet coefficient or detail images. In Fig. 3a, x[n] represents the original image; in both Fig. 3a and 3b: A, HD, VD and DD are the sub-images produced after one level of transformation. The A sub-image is the approximation image and results from applying the scaling or low-pass filter to both rows and columns. A subsequent level of transformation would be applied only to this sub-image. The HD sub-image contains the horizontal details (from low-pass on rows, highpass on columns), the VD sub-image contains the vertical details (from high-pass on rows, lows-pass on columns) and the DD sub-image contains the diagonal details (from high-pass, or wavelet filter, on both rows and columns). Drawback of decimated algorithm is not shift-invariant, which means that it is sensitive to shifts of the input image. The decimation process also has a negative impact on the linear continuity of spatial features that do not have a horizontal or vertical orientation. These two factors tend to introduce artifacts when the algorithm is used in applications such as image fusion Undecimated algorithm: Shift-variance is caused by the decimation process, and can be resolved by using the undecimated algorithm. The undecimated algorithm addresses the issue of shift-invariance. It does so by suppressing the down-sampling step of the decimated algorithm and instead up-sampling the filters by inserting zeros between the filter coefficients. Algorithms in which the filter is up-sampled are called à trous, meaning with holes [25][30].As with the decimated algorithm, the filters are applied first to the rows and then to the columns. In this case, however, although the four images produced (one approximation and three detail images) are at half the resolution of the original image; they are the same size as the original image. The approximation images from undecimated algorithm represented as, the spatial resolution

6 JAGALINGAM P., ARKAL VITTAL HEGDE becoming coarser at each higher level and the size as the original image. The undecimated algorithm is redundant, meaning some detail information may be retained in adjacent levels of transformation. It also requires more space to store the results of each level of transformation and, although it is shift-invariant, it does not resolve the problem of feature orientation. Shift-invariance is necessary in order to compare and combine wavelet coefficient images. Without shift invariance, slight shifts in the input signal will produce variations in the wavelet coefficients that might introduce artefacts in the fused image. Shiftvariance is caused by the decimation process, and can be resolved by using the undecimated algorithm. However, the other problem with standard discrete wavelet transforms is the poor directional selectivity, meaning poor representation of features with orientations that are not horizontal or vertical, which is a result of separate filtering in these directions[26][30] Non-separated: One approach for dealing with shift variance is to use a non-separated, two-dimensional wavelet filter derived from the scaling function. This produces only two images, one approximation image, and one detail image, called the wavelet plane. The wavelet plane is computed as the difference between the original and the approximation images and contains all the detail lost as a result of the wavelet decomposition. As with the undecimated DWT, a coarser approximation is achieved by up-sampling the filter at each level of decomposition; correspondingly, the filter is downsampled at each level of reconstruction. Some redundancy between adjacent levels of decomposition is possible in this approach, but since it is not decimated, it is shift-invariant, and since it does not involve separate filtering in the horizontal and vertical directions, it better preserves feature orientation [27]. 6.3 The dual-tree complex wavelet transforms (DT-CWT): The Major problem with standard discrete wavelet transforms is the poor directional selectivity, meaning poor representation of features with orientations that are not horizontal or vertical, which is a result of separate filtering in these directions. This Problem leads to the development of [34] dual-tree complex wavelet transform (DT-CWT) is an over-complete wavelet transform that provides both good shift invariance and directional selectivity over the DWT, although there is an increased memory and computational cost. Two fully decimated trees are produced, one for the odd samples and one for the even samples generated at the first level. The DT- CWT has reduced over completeness compared with the SIDWT, an increased directional sensitivity over the DWT and is able to distinguish between positive and negative orientations. The DT-CWT gives perfect reconstruction as the filters are chosen from a perfect reconstruction bi-orthogonal set. It is applied to images by separable complex filtering in two dimensions. The bi-orthogonal Antonini and Q-shift filters are used. 6.4 Contourlet Transform: Contourlet transform is a new image analysis tool, which is anisotropic and has good directional selectivity [40]. So it can accurately represent the image edges information in different scale and different direction frequency sub-bands, and some fusion algorithms based on the Contourlet transform have been proposed in recent years. 2 - D separate wavelet is good at isolating the discontinuities at object edges, but it can only capture limited directional information. Contourlet transform can effectively overcome the disadvantages of wavelet. Contourlet transform is a multi-scale and multidirection framework of discrete image. In the transform, the multiscale analysis and the multidirection analysis are separated in a serial way. The Laplacian pyramid [15][17] is first used to capture the point discontinuities, and then followed by a directional filter bank (DFB) to link point discontinuities into linear structures. The combination of a Laplacian pyramid and a directional filter bank is a double filter bank structure.the basis function of contourlet transform has 2 l directions and flexible ratio of length to width. Contourlet can decompose image into 2 l directional subbands, which means that the directions of each level are arbitrary and contourlet is anisotropy. So contourlet can achieve the optimal approximation rate for representing any 2-D piecewise smooth contours. 6.5 Nonsubsampled contourlet transform: The contourlet transform is a multidirectional and multiscale transform that is constructed by combining the Laplacian pyramid [15][17] with the directional filter bank (DFB). The pyramidal filter bank structure of the contourlet transform has very little redundancy. However, designing good filters for the contourlet transfom is a difficult task. In addition, due to downsamplers and upsamplers present in both the Laplacian pyramid and the DFB, the contourlet transform is not shift-invariant. [43] proposed an overcomplete transform that we call the nonsubsampled contourlet transform (NSCT). Transform divided into two shift-invariant parts: 1) a nonsubsampled pyramid structure that ensures the multiscale property and 2) a nonsubsampled DFB structure that gives directionality. The combinations of these two can preserve more details in source images and further improve the quality of fused image. The multiscale property of the NSCT is obtained from a shift-invariant filtering structure that achieves subband decomposition similar to that of the Laplacian pyramid. This is achieved by using twochannel nonsubsampled 2-D filter banks. The directional filter bank of Bamberger and Smith [18] is constructed by combining critically-sampled twochannel fan filter banks and resampling operations.

7 Pixel Level Image Fusion A Review on various techniques The result is a tree-structured filter bank that splits the 2-D frequency plane into directional edges. A shiftinvariant directional expansion is obtained with a nonsubsampled DFB (NSDFB). Main motivation is to construct a flexible and efficient transform targeting applications where redundancy is not a major issue.the NSCT is a fully shift-invariant, multiscale, and multidirection expansion that has a fast implementation. The design problem is much less constrained than that of contourlets. This enables NSCT to design filters with better frequency selectivity there by achieving better sub band decomposition. NSCT provide a framework for filter design that ensures good frequency localization in addition to having a fast implementation through ladders steps. The NSCT has proven to be very efficient. All these multiscale transform based fusion methods need to perform the following Steps: 1. Perform the forward transform on the source images to obtain their multiscale representations (transform coefficients with different scales and directions). 2. Combine these multiscale representations to obtain the fused multiscale coefficients, according to the fusion rules designed for certain purposes. 3. Perform the inverse-transform over the combined multiscale coefficients to obtain the fused result. 7. Conclusions: This paper explores extensively the literature of image fusion techniques. It is clear that Traditional image fusion technique such as PCA, IHS and Brovey transform produce color distortion problem as major one. To overcome this problem researchers developed multiscale transforms such as pyramid method, wavelet transform, discrete wavelet transform, contourlet transform etc Transforms method is largely characterized by the choice of filters, transform coefficient; as a result it is required to select the appropriate filters and transform coefficient for each application case in order to represent salient features of the source images. However, each multiscale transform has its own merits and demerits. For example, wavelets, contour transform and discrete complex wavelet transform (DWCT) bases are appropriate to reveal image details, lines and periodic textures respectively, but none of them can simultaneously capture two or more features. Therefore, there is no single transform which is optimal to completely represent all features because the content of an image is often complex and contains structures with different features. Moreover, the completeness and effectiveness of the transformed representations for the underlying information of the source images are crucial to the fusion quality. Future researchers can base their studies on the framework to develop the image fusion algorithm to effectively and completely represent the source images. 8. References: [1] RickS.Blum and Zheng Liu, Multi-sensor Image fusion and its applications, Taylor & Francis Group, LLC, [2] T.Stathaki, Image fusion: Algorithms and Applications. New York: Academic, [3] W.J Carper, T.M.Lillesand and R.W. Kiefer, The use of Intensity Hue Saturation transform for merging SPOT panchromatic and multispectral image data, Photogramm.Eng. Remote Sens., vol.60, no.11, pp , [4] T. Tu, S. Su, H.C. Shyu, P.S. Huang, A new look at IHS-like image fusion methods, Information Fusion pp [5] Shan-long LU, Le-jun ZOU, Xiao-hua SHEN, Wenyuan WU, Wei ZHANG, Multi-spectral remote sensing image enhancement method based on PCA and IHS transformations journal of Zhejiang Univ- Sci A (Appl Phys & Eng) vol12 no.6, pp , 2011 [6] P.S. Chavez Jr., S.C. Sides, J.A. Andeson, Comparison of three different methods to merge multi resolution and multispectral data: landsat TM and SPOT panchromatic, Photogrammetric Engineering & Remote Sensing 57 (1991) [7] P.S Chavez and A.Y.Kwarteng, Extracting spectral contrast in landsat Thematic Mapper image data using selective principle component analysis, Photogramm.Eng.RemoteSens.,vol.55,no. 3,pp ,1989. [8] KANG TingJun, ZHANG XinChang, WANG HaiYing, Assessment of the fused image of multispectral and panchromatic images of SPOT5 in the investigation of geological hazards, Science in China Series E: Technological Sciences, vol. 51, Supp. II, pp [9] A.R. Gillespie, A.B. Kahle, R.E. Walker, Color enhancement of highly correlated images. II. Channel ratio and chromaticity transformation techniques, Remote Sensing of Environment vol, 22 pp [10] B.J.Burt, The Pyramid as a structure for Efficient Computation, Multiresolution Image Processing and Analysis, London: Springer- Verlag, pp. 6-35, [11] A.Toet, morphological pyramidal image decomposition, pattern recognition letters, 9 pp , [12] Alexander Toet, Hierarchical Image Fusion, Machine Vision and Application [13] A.Toet, L.J.van Ruyven,J.M. Valeton, Merging thermal and visual images by a contrast pyramid, Optical Engineering.,vol.28 no.7, pp , July [14] A.Toet, Image fusion by a ratio of low-pass pyramid, Pattern Recognition Letters, vol.9, pp , [15] P.J. Burt, E.H. Adelson, The laplacian pyramid as a compact image code, IEEE Transactions on Communications 31 (1983) [16] E.H.Adelson, C.H. Anderson, J.R Bergen, P.J Burt, J.M Ogden, Pyramid methods in image processing, RCA Corporation, pp.33-41, 1984.

8 JAGALINGAM P., ARKAL VITTAL HEGDE [17] Wencheng Wang, Faliang chang, A multi focus Image Fusion method based on laplacian Pyramid, Journals of Computers., vol.6, no.12, pp , December [18] Luciano Alparone, Bruno Aiazzai, Stefano baronti, Andrea Gaezelli, Multispectral and panchromatic data fusion assessment without reference, photogrammetric Engineering & Remote Sensing, vol.74, no.2, pp , Feb [19] Konstantinos G. Derpains, The Gaussian Pyramid, Feb 5, [20] Valdimir S. petrovic, Costas S.Xydeas, Gradient Based Multiresolution Image fusion, IEEE Transactions on Image Processing, vol.13, no.2, pp , Feb [21] Zhijun Wang, Djemel Ziou, Costas Armenakis, A comparative Analysis of Image Fusion Methods, IEEE Transactions on Geoscience and Remote Sensing, Vol.43, no.6, pp , June [22] [22]Leila Fonseca, Laercio Namikawa, Emiliano Castejon, Lino Carvalho, Carolina Pinho, Aylton Pagamisse, Image Fusion for remote sensing applications, pp [23] Israa Amro, Javier Mateos, Miguel Vega, Rafael Molina, Aggelos K Katsaggelos, A survey of classical methods and new trends in pansharping of multispectral images, EURASIP Journal on advances in Signal processing, 2011, 2011:79 [24] Dr.Nikolaos Mitianoudis, Image fusion: theory and application, [25] Mallat, S.G., A Wavelet Tour of Signal Processing, second ed. Academic Press, San Diego. [26] Kingsbury, N.G., Image processing with complex wavelets. Philosophical Transactions of the Royal Society of London, Series A: Mathematical, Physical & Engineering Sciences 357 (1760), [27] Gonzalez-Audicana, M., Otazu, X., Fors, O., Seco, A., Comparison between Mallat's and the à trous discrete wavelet transform based algorithms for the fusion of multispectral and panchromatic images. International Journal of Remote Sensing 26 (3), [28] G.pajares and J.M. Cruz, A Wavelet-based image fusion tutorial, Pattern Recognit., vol.37, no.9, pp , [29] F.sadjadi, Comparative image fusion analysis, in Proc.IEEE Conf.Comput.Vision Pattern Recogn, vol.3, [30] Krista Amolins, Yun Zhang, and Peter Dare, Wavelet based image fusion techniques- An introduction, review and comparison, ISPRS Journal of Photogrammetric and Remote sensing, Vol.62, pp , [31] J.Nunez, X.Otazu, O.Fors, A.Prades, V.Pala, and Roman Arbiol, Multiresolution based image fusion with additive wavelet decomposition, IEEE Trans.Geosci.Remote sensing, vol.37, pp , [32] K.A. Kalpoma and j.kudoh, Image fusion processing for IKONOS 1-m color imagery, IEEE Trans. on Geosci. Remote Sens., vol.45, no.10, pp , oct [33] RC Gonzalez and RE Woods, Digital Image processing, 2 nd Ed., Englewood Cliffs, NJ: Prentice- Hall, Inc., [34] J.E.Fowler, The redundant discrete wavelet transform and additive noise, IEEE signals processing letters, Vol.12, No.9, [35] V.Vijayaraj, N.H Younan and C.G.O Hara, Quantitative analysis of pansharpened images, Optical Engineering, Vol.45, no.4, [36] Pushkar Pradham, Nicolas H. Younan and Roger L.King, Concepts of image fusion in remote sensing applications, Mississippi state university, USA. [37] K.Amolins, Y.Zhang, and P.Dare, Wavelet based image fusion techniques An introduction, review and comparisons, ISPRS J.Photogramm.Remote Sens., vol.62, no.4, pp , sep [38] Hui Li, B. S Manjunath, Smjit K. Mitra,, A Contour-Based Approach to Multisensor Image Registration, IEEE Transactions On Image Processing, vol.4, no. 3,pp ,March [39] HAI-HUI WANG, A New Multiwavelet-Based Approach to Image Fusion, Journal of Mathematical Imaging and Vision 21:pp , [40] Miloud Chikr El-Mezouar, Kidiyo Kpalma, Nasreddine Taleb, Joseph Ronsin, A Pan- Sharpening Based on the Non-Subsampled Contourlet Transform: Application to Worldview-2 Imagery IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing. [41] S. Yang, M. Wang, L. Jiao, R. Wu, and Z. Wang, Image fusion based on a new contourlet packet, Inf. Fusion, vol. 11, no. 2, pp , [42] Q. Zhang and B. L. Guo, Multifocus image fusion using the nonsubsampled contourlet transform, Signal Process, vol. 89, no. 7, pp , [43] Y. Chai, H. Li, and X. Zhang, Multifocus image fusion based on features contrast of multiscale products in nonsubsampled contourlet transform domain, Optik, vol. 123, pp , [44] Y. Chai, H. Li, and X. Zhang, Multifocus image fusion based on features contrast of multiscale products in nonsubsampled contourlet transform domain, Optik Int. J. Light Electron Opt., vol. 123, no. 7, pp , [45] Gaurav Bhatnagar, Q.M. JonathanWu, Zheng Liu, Directive Contrast Based Multimodal Medical Image Fusion in NSCT Domain, IEEE Transactions on Multimedia, vol. 15, no. 5, pp August [46] Z.Wang, D.Ziou, C.Armenakis and D.Li, A comparative analysis of image fusion methods, IEEE.trans, Geos. Remote sens., vol.43, pp , 2005.