123 20106 JOURAL OF GEO2IFORATIO SC IECE Vol112, o13 Jun1, 2010, 3 (, 350108;, 350108) :,, : allat (DW T) trous ( SW T) (SCT),IHS PCA IKOOS,,,,, DW T SW T SCT ; DW T SW T SCT IHS PCA,, IHS PCA,, SCT PCA, : ; ; ; 1,,, IKOOS,,, IHS [ 1 ] PCA [ 2 ], [ 3 ],,,,,,, allattrous allat,,, ( Decimated discrete wavelet transform, DW T) [ 4 ] trous,, ( Stationary wavelet transform, SW T) [ 5 ] 3, Cunha(2006) (onsubsamp led contourlet transform, SCT) [ 6 ], [ 7 ],IHS PCA,,,Gonzaglez(2004) DW T SW T IHS PCA,SPOT [ 8 ] ; Zhang ( 2005)DW TIHS,IKOOSQuickB ird [ 9 ] ; Yang(2008)SCTIHS,SPOTT [ 10 ] IHS PCA, : 2010-01 - 25; : 2010-04 - 15. : (JK2009004) ;( 706037) : (1982 - ),, E2mail: wxwfznu@163. com 3 :(1955 - ),,, E2mail: hxu@ fzu. edu. cn
420 2010,,,, DW T SW TSCT, IHS PCA (DW T2I SW T2 SCT2I DW T2P SW T2P SCT2P) [ 8-11 ], IKO2 OS,,,, 2 2. 1 (1) IHSPCA IHSPCA Pan IHS ( I) ( PC1),,, Pan,, [ 12 ] (2) DW T SW TSCT DW T SW TSCT (1) DW T, 1 3 DW T, 1 /4 SW TSCT,,, DW T SW T 3,SCT 2 [ 4, 6 ] DW T SW SCT, ( Pan) (DW T SW T SCT), Pan,, DW T SW TSCT 1DW T SW SCT (HD: : VD: ; DD:) Fig. 1ulti2resolution decomposition by DW T, SW T and SCT, respectively. (HD: horizontal details; VD: vertical details; DD: diagonal details) (3) DW T SW SCTIHS PCA DW T SW TSCT IHS PCA, IHS PCA,Pan,DW T SW TSCTPan I (PC1), Pan I (PC1), DW T SW TSCTI ( PC1), IHS (PCA ) 2. 2,, [ 13 ], (1) RSE (Rootean Square Error), RSE [ 14 ] :
3 : 421 R S E i = m =1 n =1 [ F i (m, n) - B i (m, n) ] 2 / ( ) (1),, m n, F, B, i,, (2) CORR i = (B m =1 n =1 i (m, n) - B i ) ( P (B m =1 n =1, ( Pan),, Pan,Lap lacian,, [ 15 ] : (m, n) - P) (B i (m, n) - B, B P Lap lacian, B i Lap lacian i; B ib i ; P P ; B B CORR, 3 IKOOS,, m = n =1 i ) m =1 n =1 i (m, n) - B i ) ( P (m, n) - B ) (2) ( P (m, n) - P), (2), 596 596 1m, 526929nm; 4m, : ( 1) 445516nm ( 2) 506595nm ( 3) 632 698nm ( 4) 757853nm 02047,, 2 1 ( a: 3 2 1 ; b: ) 2 ( c: 3 2 1 ; d: ) Fig. 2uilti2spectral images ( a, c; RGB: 3 2 1) and their corresponding panchromatic images ( b, d) of test area 1 ( a, b) and 2 ( c, d) 3. 1 IHS PCA DW T SW T SCT DW T - I(DW T IHS ) SW T - I( SW TIHS ) SCT - I (SCTIHS ) DW T - P (DW TPCA ), SW T - P ( SW TPCA ) SCT - P (SCT PCA )11,,, IHSCylinder DW TSW T db4, SCTpyr, dmaxflat7 IHS,,, 3 2 14 3 2IHS 3 1(RGB: 3 2 1), 2 ( a),4 2( RGB: 3 2 1), 2 ( c)
422 2010
3 : 423 3. 2 3 4:, IHSPCA, ;,,,, DW T SW TSCT,,, DW T2 P SW T2P SCT2P (1)( 2) RSE CORR,4 RSECORR (1), IHS RSE CORR, IHS (RGB: 3 2 1, RGB: 4 3 2) 1( R SE) ( CO RR) Tab. 1 ean R SE and correla tion coeff ic ien ts of or ig ina l and each fused image RSE CORR 1 2 1 2 0. 000 0. 000 0. 012 0. 038 IHS 239. 091 234. 041 0. 910 0. 902 PCA 262. 879 257. 950 0. 871 0. 859 DW T 79. 960 80. 972 0. 992 0. 991 SW T 76. 583 77. 627 0. 996 0. 996 SCT 76. 580 77. 619 0. 997 0. 997 DW T2I 54. 268 56. 090 0. 902 0. 895 SW T2I 52. 572 54. 583 0. 902 0. 898 SCT2I 52. 370 54. 332 0. 907 0. 899 DW T2P 49. 209 49. 766 0. 864 0. 853 SW T2P 47. 774 48. 501 0. 868 0. 856 SCT2P 47. 563 48. 239 0. 868 0. 856 (1) RSE,,, 1,PCAIHS,RSE 230, PCA IHSPan IHS,,Pan DW T SW T SCTRSE70 80,IHS PCA,, Pan, DW T2I SW T2I SCT2 I DW T2P SW T2P SCT2P RSE 60, DW T SW SCT,IHS PCA DW T SW SCTIHS PCA,, DW T SW SCT,,, Pan DW TSW TSCT, DW T2PSW T2P SCT2P,RSE DW T2I SW T2I SCT2I, PCA,, IHS, IHS,,Pan,, SCT SW T RSE, DW T, SCT SW T DW T, DW T,,SW T SCT,,,, SCT SW T (2) CORR 1,, Pan, 0. 85,, CORR DW T SW SCT, Pan,, Pan DW T2I SW T2I SCT2ICORR
424 2010, : IHS, IHSCORR IHS,,,,, DW T2P SW T2P SCT2P PCA 4,, IKOOS,, : (1)(DW T SW T SCT DW T2I SW T2ISCT2IDW T2P, SW T2PSCT2P ), IHS PCA,, DW T SW SCT IHSPCA,, SCT, SCT2P (2) DW T SW SCTIHS PCA DW T SW SCT, PCA, IHS, IHS ( 3 ), SCT SCT2I SCT2PSW T SW T2I SW T2P, SCT SCT SW T,DW T, DW TSW T SCTIHS PCA,,,, [ 1 ]Haydn R, Dalke W, Henkel J, et al. App lication of IHS Color Transform to the Processing of ultisensor Data and Image Enhancement [ C ]. Proceeding International Sympo2 sium on Remote Sensing of A rid and Sem i2a rid Lands, 1982, 559-616. [ 2 ]Chavez P S, Stuart J, Sides C, et al. Comparison of Three D ifferent ethods to erge ultiresolution and ultispec2 tral Data: Landsat T and SPOT Panchromatic [ J ]. Photo2 grammetry and Remote Sensing, 1991, 57 (3) : 259-303. [ 3 ]Dou W, Chen Y, L i X, et al. A General Framework for Component Substitution Image Fusion: An Imp lementation U sing the Fast Image Fusion ethod [ J ]. Geosciences, 2007, 33 ( 26) : 219-228. Computers and [ 4 ]allat S G. A Theory for ultiresolution Signal Decomposi2 tion: The W avelet Rep resentation [ J ]. IEEE Transactions on Pattern Analysis and achine Intelligence, 1989, 11 (7) : 674-693. [ 5 ]Lang, Guo H, Odegard J E, et al. onlinear Processing of a Shift InvariantDW T foroise Reduction [ R ]. Interna2 tional Society for Op tical Engineering on W avelet App lica2 tions, 1995, 2491. [ 6 ]Cunha A L, Zhou J, Do. The onsubsamp led Cont2 ourlet Transform: Theory, Design, and App lications [ J ]. IEEE Transactions on Image Processing, 2006, 15 ( 10 ) : 3089-3101. [ 7 ],,. Contourlet [ J ]., 2007, 35 (10) : 1934-1938. [ 8 ] Gonzaglez, Saleta L J, CatalagR G, et al. Fusion of ultispectral and Panchromatic Images U sing Imp roved IHS and PCA ergers Based on W avelet Decomposition [ J ]. IEEE Transactions on Geoscience and Remote Sensing, 2004, 42 (6) : 1291-1299. [ 9 ] Zhang Y, Hong G. An IHS and W avelet Integrated Ap2 p roach to Imp rove Pan - sharpening V isual Quality of atu2 ral Colour IKOOS and QuickB ird Images [ J ]. Fusion, 2005, 6 (3) : 225-234. Information [ 10 ] Yang X, J iao L. Fusion A lgorithm for Remote Sensing Im2 ages Based on onsubsamp led Contourlet Transform [ J ]. Acta Automatica Sinica, 2008, 34 (3) : 274-281. [ 11 ]Q iang Z, Guo B. ultifocus Image Fusion U sing the on2 subsamp led Contourlet Transform [ J ]. 2009, 89 (7) : 1334-1346. Signal Processing, [ 12 ]W ang Z, Ziou D, A rmenakis C, et al. A Comparative A2 nalysis of Image Fusion ethods [ J ]. IEEE Transactions on Geoscience and Remote Sensing, 2005, 43 ( 6) : 1391
3 : 425-1401. [ 13 ],. ATLAB IHS [ J ]., 2008, 10 (5) : 670-677. [ 14 ]L iu Z, Forsyth D, Laganigere R. A Feature2based etric for the Quantitative Evaluation of Pixel2level Image Fusion [ J ]. Computer V ision and Image Understanding, 2008, 109 (1) : 56-68. [ 15 ]. Landsat 7 ET + [ J ]., 2005, 9 (2) : 186-194. Com par ison am ong ulti2resolution D ecom position2ba sed Image Fusion ethods WU Xuewen, XU Hanqiu ( Institute of Rem ote Sensing Inform ation Engineering, Fuzhou U niversity, Fuzhou 350108, China; College of Environm ent and Resources, Fuzhou U niversity, Fuzhou 350108, China) Abstract: It is of great value to fuse a high2resolution panchromatic image and low2resolution multi2spectral images for object recognition. W ith the theory of multi2resolution analysis app lied in the field of image p rocessing, there e2 merged m any kinds of multi2resolution decomposition2based fusion methods, such as allat wavelet transform (DW T) 2based, trous wavelet transform ( SW T) 2based and nonsubsamp led contourlet transform (SCT) 2based fusion methods, or the hybrid of them. The hybrid methods include DW T2I ( integrate the DW T with IHS), SW T2 I ( integrate the SW T with IHS), SCT2I ( integrate the SCT with IHS), DW T2P ( integrate the DW T with PCA), SW T2P ( integrate the SW T with PCA ) and SCT2P ( integrate the SCT with PCA ). These fusion meth2 ods were emp loyed in two subsets from an IKOOS image, rep resenting different land cover types. One subset is mainly covered by building, the other is covered by vegetation, water and building. The size of each subset image is 596 596 p ixels in panchromatic band. The performance of each fusion m ethod has been further quantitatively analyzed. The m ean R SE ( Root ean Square Error) and correlation coefficients of original and each fused image were used to m easure spectral fidelity and high spatial frequency information gain. The study show s that all multi2 resolution decomposition2based m ethods have high spatial frequency information gain, in which the DW T, SW T, SCT methods are better. PCA 2based hybrid m ethods have better spectral fidelity than IHS2based hybrid methods, but have lower high spatial frequency information gain. A ll hybrid methods have m uch better spectral fidelity than the DW T, SW T, SCT methods and original IHS, PCA methods. The SCT2based hybrid methods are close to the SW T2based hybrid m ethods in both spectral fidelity and high spatial frequency information gain. better than DW T2based hybrid methods. The SCT2P is the best in spectral fidelity. They are much Key words:m ulti2resolution decomposition; image fusion; wavelet transform; contourlet transform