Spectral information analysis of image fusion data for remote sensing applications

Transcription

1 Geocarto International ISSN: (Print) (Online) Journal homepage: Spectral information analysis of image fusion data for remote sensing applications Yuhendra Yusuf, Josaphat Tetuko Sri Sumantyo & Hiroaki Kuze To cite this article: Yuhendra Yusuf, Josaphat Tetuko Sri Sumantyo & Hiroaki Kuze (2013) Spectral information analysis of image fusion data for remote sensing applications, Geocarto International, 28:4, , DOI: / To link to this article: Accepted author version posted online: 10 May Published online: 31 May Submit your article to this journal Article views: 532 View related articles Citing articles: 3 View citing articles Full Terms & Conditions of access and use can be found at Download by: [Shanghai Jiao Tong University School of Medicine] Date: 28 January 2016, At: 11:33

2 Geocarto International, 2013 Vol. 28, No. 4, , Spectral information analysis of image fusion data for remote sensing applications Yuhendra Yusuf a,b*, Josaphat Tetuko Sri Sumantyo a and Hiroaki Kuze a a Center for Environmental Remote Sensing (CEReS), Chiba University, Chiba, Japan; b Department of Informatics, Faculty of Engineering, Padang Institute of Technology, Padang, Indonesia (Received 21 December 2011; final version received 7 May 2012) Fusion of images with different spatial resolutions has the capability of improving visualization and spatial resolution and enhancing structural/textural information of the involved images, while preserving the spectral information in multi-spectral (MS) images. In this paper, various fusion methods have been examined in the data fusion of GeoEye-1 and QuickBird imagery, followed by subsequent image control. The effectiveness of five techniques, the Gram-Schmidt (GS), high-pass filtering (HPF), modified intensity-hue-saturation (M-IHS), fast Fourier transform (FFT)-enhanced IHS (FFT-E) and wavelet principal component analysis (W-PCA), has been evaluated through visual inspection, histogram analysis and correlation analysis. Also, image quality information is assessed by means of global spectral information (relative dimensionless global error, relative average spectral error), spectral distortion (peak signal-to-noise ratio), spectral (bias, root means square error) and spatial information (mean, standard deviation) of the fused images. In addition, the extraction of object boundary is tested and evaluated using Canny edge detection. The results show that most of the image fusion techniques preserve spectral information of original image, but occasionally with some spectral distortion. It has been found that the GS method, followed by HPF, yields the best information quality in the fused image, suitable for improving visual interpretation and data quality from the viewpoint of remote sensing applications. Keywords: image fusion; spatial resolution; colour preservation; information quality 1. Introduction Satellite remote sensing (RS) offers a wide variety of image data with different characteristics. Temporal, spatial, radiometric and spectral resolution should be chosen so as to match the requirements for each of the various applications (Al- Wassai et al. 2011). Recently, high-resolution RS data are considered indispensable for monitoring important aspects on the earth s surfaces. Nowadays more and more data are provided from various types of satellite sensors such as IKONOS, SPOT-5, WordView-2, QuickBird, GeoEye, Orbview, etc. Most of these sensors have multispectral (MS) images, which yield spectral information with relatively low spatial *Corresponding author. yuhendra_st@yahoo.com Ó 2013 Taylor & Francis

3 2922 Y. Yusuf et al. resolution. In contrast, the panchromatic (PAN) image is a greyscale image, but its wide spectral coverage makes it possible to attain much higher spatial resolution (Zhang 2002). Generally, a PAN band covers a broad wavelength range of the visible (VIS) and near infrared (NIR) spectrum, while each of the MS bands covers just a narrow spectral range. The high spatial information content of PAN images is wellsuited for intermediate scale mapping and urban classification studies (for example, Konstantinos and Nikolakopoulos 2008, Cetin and Musaoglu 2009). Although the combinatory analysis of PAN and MS images can, in principle, provide all the spatial and spectral features concerning the target pixels, better interpretation of high-resolution satellite imagery can often be attained using a PAN-sharpened image that is obtained through image fusion (Vijayaraj et al. 2006, Kumar et al. 2009). Thus, multi-sensor data fusion has become an important issue in image processing, since merging the data of high spatial resolution (PAN) with those of low spatial resolution (MS) allows one to produce imagery with higher quality more suitable for subsequent applications such as image processing and classification (Andrea and Filippo 2007, Dou et al. 2007). So far many researchers have addressed the problem of multi-resolution image fusion for RS applications, proposing different pan-sharpening methods (Pohl and Van Genderen 1998). These methods are based on some form of image transformation such as intensity-hue-saturation (IHS) (Choi 2006, Karathanassi et al. 2007), colour normalized (CN), Brovey (Du et al. 2007, Bovolo et al. 2010), and principal component analysis (PCA) (Shah et al. 2008). Other methods such as Gram-Schmidt (GS) (Kumar et al. 2009), synthetic variable ratio (SVR) (Wang et al. 2008, Rahman et al. 2010), Ehler (Ehler et al. 2010), and high-pass filtering (HPF) (Wald et al. 1997) rely on the intensity modulation. In addition, several researchers have proposed the use of wavelet transform (Shi et al. 2005, Acerbi et al. 2006) or discrete wavelet transform (Li et al. 2005) to extract geometric edge information from PAN images. The IHS, CN-Brovey, and CN-spectral techniques are the most commonly used algorithms in RS applications. However, the problem associated with the application of these fusion methods is the colour distortion that appears in analysed areas in general (Kalpoma and Kudoh 2007), and also, in the fusion of high-resolution satellite images (Zhang 2002). In this study, we apply image fusion to high-resolution images taken with GeoEye-1 and QuickBird satellites. While their PAN data have a high spatial resolution of 0.5 m and 0.7 m, respectively, four reflective MS bands have a lower resolution of 2 m and 2.8 m, respectively. Table 1 lists the spectral and spatial resolution information of both satellites. The main purpose of this study is to improve the spatial resolution and enhance structural/textural information, while Table 1. Spectral range of different PAN sensors. Sensor name Spectral wavelength (mm), PAN Spatial resolution (m) GeoEye QuickBird Ikonos WorldView Spot EO1(ALI) ALOS

4 Geocarto International 2933 preserving the spectral information provided in MS images. Also, we apply the following statistical approaches for standardizing and automating the evaluation process of the PAN-sharpened images. First, qualitative assessment is made with conventional visual inspection. Then, we examine the spectral correlation between the original MS and the fused images by means of statistical parameters including histograms of various frequency bands. Finally, we apply the edge detection test, in which fused images are compared from the viewpoint of the preservation of edge information. 2. Satellite data and methods 2.1 Geo-Eye1 and QuickBird satellite data In the present work, we analyse two datasets from two different sensors onboard GeoEye-1 and QuickBird satellites. GeoEye-1 satellite was launched by GeoEye, Inc. on 6 September The area of the image used in this study is the Hobart area ( E, S) in Tasmania, Australia, obtained in December 2009 and provided by GeoEye Inc. ( Geoeye.com). The scene covers an area that includes bare soil, vegetation, and suburban residential areas with streets. QuickBird satellite, on the other hand, was launched in 2001 by DigitalGlobe, Inc. The data location is the Kolkata area ( E, N), India, obtained in November 2004 and distributed in the Global Land Cover Facility ( The PAN sensor collects information in the VIS and NIR wavelengths of mm. The area covered in this imagery is mainly an urban area with structured road, water, roof, tree, shadow and grass. The spectral bands and spectral response curves of these two sensors are summarized in Table 1 and Figure Spectral response and resolution Significant spectral distortion in image fusion occurs due mainly to the wavelength extension of the new satellite PAN sensors. Table 1 shows the spectral and spatial resolution of satellites ranges of different PAN sensors. In image fusion techniques, it is important to include sensor spectral response information (Otazu et al. 2005), since, first, in order to preserve physical meaning of merged spectral bands, the sensor spectral response for each band has to be taken into account and second, image fusion techniques aim at recovering the image obtained by an ideal virtual sensor with the same spectral sensitivity of the MS and spatial resolution of the PAN sensor. In image fusion, spatial resolution ratio plays an important role, since this parameter describes the ratio between the spatial resolution of the high-resolution PAN image and that of the low-resolution MS image (Ling et al. 2008). As summarized in Table 2, the spatial resolution ratio is 1:4 for both GeoEye-1 and QuickBird images. 2.3 Pre-processing Band selection model by the OIF Band selection is a key step in fusion techniques. For this purpose, values of optimum index factor (OIF) are useful for designating the most favourable band

5 294 4 Y. Yusuf et al. Figure 1. Table 2. Spatial resolution ratio for images fusion. Data Fused images Input images Resolution ratio I II Spectral response curves for GeoEye-1 and QuickBird satellites. GeoEye-1 PAN þ GeoEye-1 MS QuickBird PAN þ QuickBird MS GeEye-1 PAN (0.5m) þ GeoEye-1 MS (2m QuickBird PAN (0.7m) þ QuickBird MS (2.8m) 0.5:2 (1:4) 0.7: 2.8 (1:4) combination according to their information (Jensen 2005, Tsagaris and Anastassopoulus 2007). Generally, a larger standard deviation (STD) of an image infers that it involves more information. Thus, the OIF is defined (Chavez et al. 1982) as OIF ¼ X3 i¼1 s i, X 3 j¼1 r j ð1þ where s i is the STD of each of the three selected bands and r j is the correlation coefficients (CCs) between any pair formed by these bands (Table 3).

6 Geocarto International Re-sampling The most important prerequisite for accurate data fusion is precise geometric correction. The GeoEye-1 image has been geometrically corrected and registered to WGS 84 datum with the Universal Transverse Mercator (UTM) zone 45 projection. A QuickBird image also has been geometrically corrected and registered to WGS 84 datum with the UTM zone 55S projection. Since the PAN and MS images were taken at the same time with the same sensor, data fusion can be carried out directly without further registration (Xue 2005). Next, we apply re-sampling, in which each data point (pixel) in the high-resolution base map is assigned with a value based on the MS image pixels using the nearest neighbour technique. The role of this resampling is to reduce and increase the size of an image before or during the fusion process. This technique is a pre-requisite before performing image fusion process (Pohl and Van Genderen 1998). The images after re-sampling are shown in Figure 2. Table 3. OIF index for bands combination the original MS images. Satellite sensor Band combination S s i S r j OIF GeoEye QuickBird Figure 2. Pixel-level re-sampling for GeoEye-1 and QuickBird data.

7 296 6 Y. Yusuf et al. 2.4 Fusion techniques In the present analysis of GeoEye-1 and QuickBird images, pixels of PAN and MS images are fused by using an algorithm specifically developed for very highresolution satellite images. In order to improve the quality of spatial and spectral information, the following five methods are tested and compared: GS, HPF, modified HIS, FFT-enhanced (FFT-E) IHS and wavelet-pca. These methods have been chosen here, since they can yield less spectral distortion as compared with other fusion methods such as IHS, PCA, Brovey and wavelet (Te-Ming et al. 2001, 2007, Zhang 2002). In the following part of this section, we briefly describe the essential aspects of these five fusion methods Gram-Schmidt The basic concept of GS fusion technique is to simulate a high-resolution PAN band from the low-spatial MS bands with appropriate weights (Laben and Brower 2000, Kumar et al. 2009). A GS transformation is performed by employing the simulated PI as the first band. After swapping the first GS band with the high-resolution PAN image, the inverse GS transform is applied to form the pan-sharpened spectral bands High-pass filtering The HPF technique is based on injecting spatial details taken from the PAN image onto the re-sampled MS images (Schowengerdt 1998). The mathematical model can be given as F i;j ¼ MS i;j þ½pan i;j PAN i;jðw;hþ Š where F i,j is the pixel of the fused image at coordinate (i, j), MS i,,j and PAN i,,j are the corresponding pixel values in MS and PAN image, respectively; and PAN i;jðw;hþ stands for the local mean of high-resolution channel inside the window of w (width) 6 h (height) pixels centred at (i, j) Modified intensity-hue-saturation (M-IHS) In the M-IHS techniques, a high-spatial resolution PAN is combined with lowspatial resolution MS images in the following manner (Siddiqui 2003, Kumar et al. 2011):. Set the b i parameter, which denotes the relative contribution of each portion of the spectrum included in the PAN band. This is accomplished with a linear regression between the i-th MS image and the PAN image.. Set the a parameter, which is defined as a ¼ P ib i MS i 3PAN ð2þ ð3þ where MS i is the average of band i and PAN is the average of PAN band.

8 Geocarto International Generate the intensity modification ratio, r 1, by applying an RGB-to-IHS transformation on the three MS bands: r 1 ¼ a r dr þ a g dg þ a b db P i b i di ð4þ Here, a k (k ¼ r, g, b) is the numerator coefficient for red, green and blue digital number (DN) value, respectively; d k is the DN value of band used for red, green, and blue output, respectively; and d i is the DN value of band i.. Reverse the RGB-to-IHS transformation by multiplying the modification ratio (r 1 ) by the PAN band. The PAN component replaces the intensity component of IHS images, and the fused image is transformed back into the RGB colour space to generate the fused images FFT-enhanced IHS FFT-enhanced IHS transform has been developed specifically for image merging that preserves spectral characteristics (Ling et al. 2007; Ehler et al. 2010). The procedure of this method can be expressed as follows (Ling et al. 2007; Kumar et al. 2011) (see Figure 3). The re-sampled MS images are transformed from the RGB to IHS colour space to obtain the intensity (I), hue (H), and saturation (S) components, and low-pass filtering (LP) is applied to the intensity component. After HPF, the PAN image is added to the LP filtered intensity component by means of inverse fast Fourier transform (FFT 71 ). Finally, inverse IHS transformation (IHS 71 ) is performed on the IHS image to create the fused image Wavelet-principal component analysis (W-PCA) The W-PCA refers to the wavelet method in combination with PCA. The procedure of decomposition level of W-PCA method can be expressed as follows (King and Wang 2001, Gonza lez-audícana et al. 2004, Karathanassi et al. 2007); see also Figure 4:. The PAN and MS images must be co-registered. The PAN image must be resampled.. Apply the PC analysis to the MS and obtain the first principal component (PC 1 ). Figure 3. FFT-enhanced IHS flowchart (adapted from Ehlers et al. 2006).

9 298 8 Y. Yusuf et al. Figure 4. W-PCA fusion processing schemes (adapted from Gonza lez-audı cana et al. 2004).. Generate a new PAN image (PAN LL ) whose histogram matches that of the PC 1 image.. Apply the decimated wavelet decomposition to the PC 1 image and then to the corresponding PAN LL image. In the wavelet decomposition, the Daubechies four wavelet coefficients are used for each of PC 1 and PAN LL to generate a half-resolution approximation image with three wavelet coefficient images corresponding to horizontal decomposition (HD), vertical decomposition (VD) and diagonal decomposition (DD).. Inject the coefficients of the PAN image representing the spatial detail information into the PC 1 image through the inverse multi-resolution wavelet decomposition.. Apply the inverse PCA transform to the PC 1 image to obtain the fused image. 2.5 Spectral information quality of fused image The quality of fused images obtained with the above-mentioned five different fusion methods is compared in terms of the spectral information. For this purpose, here the following four approaches are chosen: qualitative (visual comparison), histogram, quantitative (statistical analysis) and edge detection. Although the visual comparison inevitably involves subjective factors and personal preference, the method can in general give grades to the visual quality of fused images (Shi et al. 2005). Nevertheless, it is desirable to give a more quantitative way to assess the quality of fused images. From the viewpoint of practical use and ease of applicability, here we employ the following seven measures to evaluate spectral similarity between the fused image and the original MS images: mean values (Shi et al. 2005), STD (Li et al.

10 Geocarto International ), bias (Acerbi et al. 2006), root mean squared error (RMSE) (Klonus and Ehlers 2007, Deshmukh and Bhosale 2010), peak signal-to-noise ratio (PSNR) (Karathanassi et al. 2007, Harish Kumar and Singh 2010), relative dimensionless global error (ERGAS) (Nencini et al. 2007, Li et al. 2010), and relative average spectral error (RASE) (Gonza lez-audícana et al. 2004; Li et al. 2010). Briefly, these parameters are defined as follows:. Mean value refers to the grey level of pixels in an image. If the image size is m 6 n, the mean value is defined as Mean ¼ 1 X m X n x ij mn i¼j j¼i. STD reflects the deviation degree of values relative to the mean of the image. Normally, the STD is defined as 1 X 1=2 STD ¼ i¼1ðms i;j MS mean Þ2 n 1 ð6þ. Bias is the difference between the mean of the original image and that of the fused image. The ideal value is zero: Bias ¼ MS mean fused mean MS mean ¼ 1 Fused mean MS mean. Relative average spectral error (RASE) is used to estimate the global spectral quality of the fused images. This index is expressed as RASE ¼ 100 M ð5þ ð7þ Xn 1 1=2 RMSE 2 ðb i Þ n ð8þ i¼1 where M is the mean radiance of the n spectral bands (B i ) of the original MS bands; RMSE is the root mean square error computed as RMSEðB i Þ¼Bias 2 ðb i ÞþSTD 2 ðb i Þ. Relative dimensionless global error in synthesis (ERGAS) was proposed by Wald (2000) as a multi-modal index to characterize the quality of process in terms of the normalized average error of each band of processed image. Increasing in ERGAS index generally indicates degradation in images due to the fusion process. ERGAS index for the fusion is expressed as follows: ð9þ ERGAS ¼ 100 dh dl 1 X 1=2 n ð10þ n i¼1 RMSE 2 Mean 2

11 Y. Yusuf et al. where d h /d l is the ratio the pixel sizes of PAN and MS (in the present case, this ratio is 1/4), and N is number of bands.. PSNR is used to reveal the radiometric distortion of the final image compared to the original image. It is widely used for the evaluation of image compression algorithms: Peak PSNR ¼ 10 log 10 RMES 2 where Peak is the maximum possible pixel value (255 for 8-bit images, 2047 for 11-bit images, and 65,535 for 16-bit images). 3. Results and discussion 3.1 Visual evaluation From the original GeoEye-1 image, colour combinations of MS bands are produced and analysed using the OIF. Only three bands (R-G-B) can be employed from the four bands. Thus, these three spectral bands should be selected so as to project the MS information as efficiently as possible onto the final pseudo-colour image, and this can be achieved by maximizing the OIF. For GeoEye-1, the highest value of average OIF (146.46) has been obtained for the band combination 2-3-4, as shown in Table 3. A slightly lower value of OIF has been obtained with the band combination of For QuickBird, the maximum OIF (130.95) has been obtained with the same band combination of Figure 5(a) shows the false-colour composite of the bands of GeoEye-1. The PAN-sharpened images resulting from the five different image fusion techniques are shown in Figure 5(b) (f), namely, (b) GS, (c) FFT-E, (d) M- IHS, (e) HPF and (f) W-PCA. A major problem in image fusion is colour distortion (Zhang 2002). Preservation of colour information is an important criterion in evaluating the performance of different image fusion methods. Before comparing with the original MS images (Figure 5(a)), the fused images must be re-sampled to match the resolution of the original MIs (Figure 2). Ideally, when a fused image is re-sampled, the original colour information should be reproduced again (Wald et al. 1997). The results in Figure 5 indicate that there is no apparent colour distortion in all the fused images. The GS fused image (Figure 5(b)) shows colour intensity slightly higher than HPF, M-IHS, FFT-E and W- PCA (Figure 5(c f)). The W-PCA result (Figure 5(f)) shows better spatial details and colour feature that are close to the original MS images, but its colour intensity is weak. Figure 6 shows the corresponding results from the QuickBird image. Concerning the colour preservation, again it has been found that all fused images are close to the original image. This means that all the techniques can preserve the spatial information for the fusion images. Particularly, GS (Figure 6(b)) and M-IHS (Figure 6(d)) exhibit more pronounced spatial information than HPF (Figure 6(c)), FFT-E (Figure 6(e)) and W-PCA (Figure 6(f)). ð11þ

12 Geocarto International Figure 5. Results of the false colour composite of the GeoEye-1 image: (a) composite of the bands of the original GeoEye-1 MS images, and the results of different fusion techniques, with (b) GS, (c) HPF, (d) M-IHS, (e) FFT-E and (f) W-PCA. Figure 6. Results of the false colour composite of the QuickBird image: (a) composite of the bands of the original GeoEye-1 MS images, and the results of different fusion techniques, with (b) GS, (c) HPF, (d) M-IHS, (e) FFT-E and (f) W-PCA.

13 Y. Yusuf et al. 3.2 Image histogram analysis The histograms of the original and fused images of GeoEye-1 and QuickBird band are shown in Figures 7 and 8, respectively. Generally, a histogram illustrates Figure 7. Histograms related to the GeoEye-1 image: (a) the grey scale distributions of the band 2 (R), 3 (G), and 4 (B) of the original MS images. Histograms for the fused images are shown for different fusion methods of (b) GS, (c) HPF, (d) M-IHS, (e) E-FFT, and (f) W- PCA.

14 Geocarto International Figure 8. Histograms related to the QuickBird image: (a) the grey scale distributions of the band 2 (R), 3 (G), and 4 (B) of the original MS images. Histograms for the fused images are shown for different fusion methods of (b) GS, (c) HPF, (d) M-IHS, (e) E-FFT, and (f) W- PCA.

15 Y. Yusuf et al Figure 9. Edge detection from the original and fused images of GeoEye-1. (a) original PAN and the result with different fusion methods of (b) GS, (c) HPF, (d) M-IHS, (e) FFT-E and (f) W-PCA. the distribution of grey scale of an image that corresponds to the statistical distribution of image brightness of image pixels. From Figure 7 (GeoEye-1), it can be seen that the histograms of fused images are very similar to those of the original

16 Geocarto International image in the case of GS, HPF, and W-PCA, while the relative band intensity has somewhat been distorted in M-IHS and FFT-E. In Figure 8, on the other hand, the histograms of band 4 (black line) show ripples, especially noteworthy for W- PCA. This is the reason for the image blur found in Figure 6. As compared with other methods, the GS-fused image exhibits better result in terms of both intensity and texture information. 3.3 Performance comparison using spectral quality A thorough examination of spectral quality is made for fused images using the following seven statistical parameters: mean, bias, STD, CCs, RMSE, ERGAS, PNSR and RASE. The fused image that best preserves the original spectral information, and hence exhibits the highest spectral quality, can be characterized with the following conditions: (i) the mean and SDT values are closer to those of the original MS images, (ii) the smallest values for the error parameters (RMSE, Bias, RASE and ERGAS), and (iii) the highest possible values of CCs and PSNR. Tables 4 11 summarize the results of the present analysis. Table 4 summarizes the CCs for the GeoEye-1 image between the original MS bands 2, 3 and 4 and the results obtained with different image fusion techniques. Here CCs are used to evaluate the spectral resemblance of two images and it can be seen from Table 4 that the resulting values are relatively high, with the average around The highest value of has been obtained in the GS-fused image. Thus, the GS method produces the result that has the least colour distortion as a whole and the best preservation of spectral characteristics. As another test, we can examine the inter-band correlation that should be preserved in image fusion. Table 5 shows the comparison of the interband correlations (B2/B3, B2/B4, and B3/B4), among the original GeoEye-1 image and between the original and fused images. The highest correlation around 0.86 is seen for the combination of Band 2 and Band 3, irrespective of the fusion method employed. The results in Table 5 show that both GS and FFT-E give values reasonably in agreement with those in the original MS images. Table 6 and Table 7 show the CCs and inter-band correlation, respectively, obtained for the QuickBird image. From CCs, it is seen that fused images based on GS and FFT-E are well correlated with the original MS images. The results of the inter-band correlation (Table 7), on the other hand, indicate Table 4. Spectral correlation for GeoEye-1 between the original MS bands 2, 3 and 4 and the fused results obtained with different techniques. Image fusion technique Band 2 ( ) Geo-Eye1 waveband (mm) Band 3 ( ) Band 4 ( ) Average MS* GS HPF M-HIS FFT-E W-PCA Note: *Original image.

17 Y. Yusuf et al. that GS fusion method gives results that show similarity with the original MS images. In Table 8, we can see that the fused image based on GS (followed by HPF) exhibits the smallest ERGAS and RASE errors when all bands are considered. Smaller values of spectral distortion given by PSNR (Table 9) are associated with high RMSE values in Table 10. For FFT-E, for example, a high RMSE value of 17.60% (band 4 of GeoEye-1) is accompanied with a small PSNR value of 8.20%, whereas for GS, we have a small RMSE error of 0.63% (band 2 of GeoEye-1) and a large PSNR value of 37.12%. On the basis of Tables 8 9, one can conclude that, as a whole, the GS method provides good spectral quality, preserving the spectral information contained in the original MS images. Table 5. images. Inter-band correlation between spectral bands for original GeoEye-1 and fused Image fusion technique Band 2 3 Band 2 4 Band 3 4 MS GS HPF M-HIS FFT-E W-PCA Table 6. Spectral correlation for QuickBird between the original MS bands 2, 3 and 4 and the fused results obtained with different techniques. QuickBird waveband (mm) Image fusion technique Band 2 ( ) Band 3 ( ) Band 4 ( ) Average MS GS HPF M-HIS FFT-E W-PCA Table 7. images. Inter-band correlation between spectral bands for original QuickBird and fused Image fusion technique Band 2 3 Band 2 4 Band 3 4 MS GS HPF M-HIS FFT-E W-PCA

18 Geocarto International Table 8. Global spectral information evaluation indicator of fused image. Parameter Fusion techniques Geo-Eye1 QuickBird Band-2 Band-3 Band-4 Band-2 Band-3 Band-4 ERGAS GS HPF M-HIS FFT-E W-PCA RASE GS HPF M-HIS FFT-E W-PCA Table 9. Parameter Evaluation of fused images with a distortion indicator (PSNR). Fusion techniques Geo-Eye1 QuickBird Band-2 Band-3 Band-4 Band-2 Band-3 Band-4 PSNR GS HPF M-HIS FFT-E W-PCA Table 10. Parameter Evaluation of fused images with spectral information indicators (bias and RMSE). Fusion techniques Geo-Eye1 QuickBird Band-2 Band-3 Band-4 Band-2 Band-3 Band-4 Bias GS HPF M-HIS FFT-E W-PCA RMSE GS HPF M-HIS FFT-E W-PCA Subsequently, the HPF method gives better results as compared with FFT-E, M- IHS and W-PCA. Finally, Table 11 shows the evaluation of spatial quality by means of mean and STD. It is seen that for both GeoEye-1 and QuickBird, the fusion techniques based on GS and HPF give values close to those of the original MS images.

19 Y. Yusuf et al. Table 11. Evaluation of fused images with spatial information indicators (mean and STD). Parameter Fusion techniques Geo-Eye1 QuickBird Band-2 Band-3 Band-4 Band-2 Band-3 Band-4 Mean MS GS HPF M-HIS FFT-E W-PCA STD MS GS HPF M-HIS FFT-E W-PCA Edge detection Edge detection is one of the most frequently used techniques in digital images processing. Edge detection refers to the process of identifying and locating sharp discontinuities in an image (Senthilkumaran and Rajesh 2009), as a precursor step to feature extraction and object segmentation. In order to examine the fusion performance, the edge detection algorithm is applied to the data fusion products. Some edge detection techniques usually adopt edge operators such as Sobel, Prewitt, Robert or Canny operators. Here, we employ the Canny edge detection, and compare the feature extraction between the fused images and the PAN image. The Canny edge detection algorithm uses an optimal edge detector based on a set of criteria which include finding most of the edges by minimizing the error rate, marking edges as closely as possible to the actual edges to maximize localization, and marking edges only once when a single edge exists for minimal response (Canny 1996). Figure 8 shows the result of the Canny edge detection applied to the original and fused GeoEy-1 images. Object boundaries detected are encircled in each panel of Figure 8. It has been found that the images produced with both the GS and HPF methods give high potential for better boundary extraction. 4. Conclusion This paper has studied the colour preservation and spectral information analysis based on four criteria, namely, visual evaluation, image histogram, spectral quality information and edge detection. In order to improve the spatial resolution and enhance structural/textural information while preserving the spectral information in MS images, five fusion methods of GS, HPF, M-IHS, FFT-E, and W-PCA have been tested, for both GeoEye-1 and QuickBird. Comparison among various fusion methods has indicated that GS and HPF techniques can lead to better results than the other three methods. The evaluation method presented in this paper will be useful for processing of RS data in the future, especially, the combinatory use of high-resolution optical images with synthetic aperture radar (SAR) images.

20 Geocarto International References Acerbi, F.W., Clevers, J.G.P.W., and Schaepman, M.E., The assessment of multi-sensor image fusion using wavelet transform for mapping the Brazalian Savana. International Journal of Applied Earth Observation and Geoinformation, 8, Al-Wassai, F.A., Kalyankar, N.V., and Al-Zaky, A.A., The statistical methods of pixelbased image fusion techniques. International Journal of Artificial Intelligence and Knowledge Discovery, 1 (3), Andrea, G. and Filippo, N., Pan sharpening of remote sensing images using a multi scale Kalman filter. Journal of Pattern Recognition Society, 40, Bovolo, F., et al., Analysis of effect of pan-sharpening in change detection. IEEE Transactions on Geoscience and Remote Sensing Letters, 7 (1), Canny, J., A computational approach to edge detection. IEEE Transactions on Pattern Analysis and Machine Intelligence, 8, Cetin, M. and Musaoglu, N., Merging hyperspectral and panchromatic image data: qualitative and quantitative analysis. International Journal of Remote Sensing, 30 (7), Chavez, P.S., Berlin, G.L., and Sowers, L.B., Statistical method for selecting Landsat MSS ratios. Journal of Applied Photographic Engineering, 8, Choi, M., A new intensity-hue-saturation fusion approach to image fusion with a tradeoff parameter. IEEE Transactions on Geosciences Remote Sensing, 44, Deshmukh, M. and Bhosale, U., Image fusion and image quality assessment of fused images. International Journal of Image Processing (IJIP), 4 (5), Dou, W., et al., A general framework for component substitution image fusion: an implementation using the fast image fusion method. Computers & Geosciences, 33, Du, Q., et al., On the performance evaluation of pan-sharpening techniques. IEEE Geosciences Remote Sensing Letters, 4 (4), Ehler, M., et al., Multi-sensor for pansharpening in remote sensing. International Journal of Image and Data Fusion, 1 (1), Ehlers, M., Greiwe, A., and Tomowski, D., On segment based image fusion. International Archives of Photogrammetry, Remote Sensing and Spatial Information Sciences, XXXVI-4/C42, Salzburg, Austria. González-Audı cana, M., et al., Fusion of multispectral and panchromatic images using improved IHS and PCA mergers based on wavelet decomposition. IEEE Transactions on Geoscience & Remote Sensing, 42, Harish Kumar, G.R. and Singh, D., Quality assessment of fused image of MODIS and PALSAR. Progress in Electromagnetics Research B, 24, Kalpoma, K.A. and Kudoh, J.I., Image fusion processing for IKONOS 1-m color imagery. IEEE Transactions on Geoscience & Remote Sensing, 45 (10), Karathanassi, V., Kolokousis, P., and Ionnidou, S., A comparison study on fusion methods using evaluation indicators. International Journal of Remote Sensing, 28 (10), King, R.L. and Wang, J., A wavelet based algorithm for pan sharpening Landsat 7 imagery. In: Geoscience and Remote Sensing Symposium, IGARSS 01, 9 13 July, Sydney, Australia, Klonus, S. and Ehlers, M., Image fusion using the Ehlers spectral characteristics preservation algorithm. GIScience & Remote Sensing, 44, Konstantinos Nikolakopoulos, G., Comparison of nine fusion techniques for very high resolution data. Photogrammetric Engineering & Remote Sensing, 74 (5), Kumar, U., et al., Comparison of 10 multi-sensor image fusion paradigms for IKONOS images. International Journal of Research and Reviews in Computer Science (IJRRCS), 2 (1), Kumar, U., Mukhopadhyay, C., and Ramachandra, T.V., Pixel based fusion using IKONOS imagery. International Journal of Recent Trend in Engineering, 1 (1), Laben, C.A. and Brower, B.V., Process for enhancing the spatial resolution of multispectral imagery using pan-sharpening [online]. United States Eastman Kodak Company (Rochester, NY), US Patent Available from: online.com/ html [Accessed 18 May 2012].

21 Y. Yusuf et al. Li, Z., et al., Color transfer based remote sensing image fusion using non-separable wavelet frame transform. Pattern Recognition Letters, 26 (13), Li, S., Li, Z., and Gong, J., Multivariate statistical of measures for assessing the quality of image fusion. International Journal of Image and Data Fusion, 1, Ling, Y., et al., FFT-enhanced IHS transform method for fusing high-resolution satellite images. ISPRS. Journal of Photogrammetry & Remote Sensing, 61, Ling, Y., et al., Effect of spatial resolution ratio in image fusion. International Journal of Remote Sensing, 29 (7), Nencini, F., et al., Remote sensing image fusion using the curvelet transform. Information Fusion, 8, Otazu, X., et al., Introduction of sensor spectral response into image fusion methods. Application to wavelet-based methods. IEEE Transactions on Geoscience & Remote Sensing, 43 (10), Pohl, C. and Van, Genderen, J.L., Multisensor image fusion in remote sensing: concepts, methods, and applications. International Journal of Remote Sensing, 19 (5), Rahman, M.M., Sri Sumantyo, J.T., and Sadek, M.F., Microwave and optical image fusion for surface and sub-surface feature mapping in Eastern Sahara. International Journal of Remote Sensing, 31 (20), Schowengerdt, R.A., Reconstruction of multispatial, multispectral image data using spatial frequency content. Photogrammetric Engineering and Remote Sensing, 46 (10), Senthilkumaran, N. and Rajesh, R., Edge detection techniques for image segmentation a survey. International Journal of Recent Trends in Engineering, 1 (2), Shah, V.P., Younan, N.H., and King, R.L., An efficient pansharpening method via a combined adaptive PCA approach and contourlets. IEEE Transactions on Geosciences Remote Sensing, 4, Shi, W., et al., Wavelet-based image fusion and quality assessment. International Journal of Applied Earth Observation and Geoinformation, 6, Siddiqui, Y The modified IHS method for fusing satellite imagery. In: Proceedings of American Society of Photogrammetry and Remote Sensing, 5 9 May, Anchorage, Alaska, Te-Ming, T., et al., A new look at HIS-like image fusion methods. Information Fusion,2, Te-Ming, T., et al., Best tradeoff for high-resolution image fusion to preserve spatial details and minimize color distortion. IEEE Transactions on Geoscience and Remote Sensing Letters, 4 (2). Tsagaris, V. and Anastassopoulus, V., Multispectral image fusion for improved RGB based on perceptual attributes. International Journal of Remote Sensing, 26 (15), Vijayaraj, V., Younan, N., and O Hara, C., Concepts of image fusion in remote sensing application. IEEE Transactions on Geosciences Remote Sensing, 10 (6), Wang, L., Cao, X., and Chen, J., ISVR: an improved synthetic variable ratio method for image fusion. Geocarto International, 23 (2), Wald, L., Ranchin, T., and Mangolini, M., Fusion of satellite images of different spatial resolutions: assessing the quality of resulting images. Photogrammetric Engineering & Remote Sensing, 63 (6), Wald, L., Quality of high resolution synthesized images: is there a simple criterion? Proceedings of International Conference on Fusion of Earth Data, 1, Xue, H.Q., Study on data fusion and classification of Landsat 7 ETM þ imagery. Journal of Remote Sensing, 9 (2), Zhang, Y., Problems in the fusion of commercial high-resolution satellite images as well as Landsat 7 images and initial solutions. International Archives of Photogrammetry and Remote Sensing (IAPRS), GeoSpatial Theory, Processing and Applications, 34(Pt,4).[CD-ROM].