An Improved Wavelet Transform Fusion Algorithm And Its Quality Evaluation

Modern remote sensing technology can provide multi-spatial,multi-spectral and multi-temporal resolution massive image data for application, but these data have their own characteristics, there are many differences between each other, so that a single data is difficult to meet the requirements of application. Remote sensing image fusion technology can efficiently combine the advantages of these raw data to complement each other, obtain a high-quality image data with high spatial resolution and multi-spectrum and provide a new way for multi-source data processing, analysis and application. At present, it is an important research topic of remote sensing image processing field.According to the information, remote sensing image fusion method can be divided into pixel level, characteristic level and decision level three layers. Domestic and overseas scholars, according to the different application, have put forward many methods, but there are some limitations. In this thesis, based on the former research, the image fusion methods at pixel level and their effect evaluation were further explored, focusing on the image fusion methods based on wavelet transformation and its improved algorithm. The main research work and results include the following aspects:Firstly, to summarize the principle of PCA, pyramid, HIS, high-pass filtering and Brovey transform algorithm, etc, perform the fusion experiments based on these common methods, and systematically compare their advantages and disadvantages, as the contrast.Secondly, to fuse the remote sensing images based on wavelet transform and its improved algorithm. Because of the multi-scale analysis characteristics of wavelet algorithm, the image fusion based on wavelet algorithm can effectively improve the serious spectral and spatial information loss phenomena of traditional methods. However, the general wavelets-based image fusion method simply replaces high frequency component of multi-spectrum data with that of pan image, which leds to the loss of much high-frequency texture information of multi-spectral data and the circular effect and ringing phenomenon because of the huge spectral differences between low-frequency information of multi-spectrum data and that of pan image The fusion method combining the traditional algorithm with wavelet transform, although improves the faults of single wavelet fusion algorithm, there is no change on fusion rules, so the effect is not good. This thesis presented a new improved wavelet algorithm. According to the characteristics of local variance algorithm that could well reflect the detail of image information, this algorithm calculated local variance of three directions high-frequency information after wavelet decomposition; through the threshold value, determined the high component of ultimate fusion image; combining with the characteristics of local differences weighted operation that could better keep low frequency information of image, determined the final low-frequency by the differences weighted operation of low-frequency information on the original two images; further used the characteristic of traditional PCA transform fusion method that could keep very good spectrum information. Then a kind of image fusion algorithm combining PCA , the wavelet transform and the local operations was formed, which radically changed the fusion rules of traditional method.Thirdly, respectively from the visual qualitative analysis, mathematical quantitative analysis and the improved degree of classification accuracy, three aspects, to comprehensively evaluate the image fusion results. The subjective comprehensive evaluation was not only based on visual qualitative analysis from color, information and spectral effect, three aspects , but also through designing certain evaluation standards for scoring and calculated the weighted average value of the scoring results of professional and non-professional people to verify the effectiveness of the subjective evaluation. The quantitative analysis based on statistics mainly from brightness information, space information and spectral information conducted an all-round quantitative evaluation on image fusion results. The evaluation based on the improved degree of classification accuracy oriented to application verified the classification accuracy improvement by image fusion algorithm, through the classification precision evaluate respectively before and after fusion. which made the fusion image directly face the application.Finally, using the Matlab software as development platform, the system which can perform the above methods of image fusion and effect evaluation to finish the case study was developed.Experiment results showed that from the qualitative analysis based on visual effect and quantitative analysis based on mathematical statistics, we could see, among the traditional pixel level image fusion methods, spectrum information retained effect of PCA transform method was the best, but spatial information loss was seriouse. To a certain extent, though wavelet transform image fusion method could improve the shortcomings of traditional methods, enhance the spatial texture information and keep the spectral information effectively, with increase of decomposition layers, fusion image would produce plaques effect, and requied running time and space also increase. The fusion method combining wavelet transform with the PCA or HIS could improved the faults of single wavelet transform image fusion algorithm, but the effect was still poor. The fusion effect of improved method combining wavelet, PCA transform and local algorithm no matter in visual effect or in quantitative analysis was better than that of other methods, which efficiently realized Pan-image and multi-spectrum data fusion. From the perspective of classification accuracy evaluation, relative to the original image, the classification precision of fusion images were all improved, among them, the classification accuracy of the fusion image based on the new method combining PCA, local algorithm with wavelet transform was the highest, relative to the primitive image, the overall accuracy was increased by 10%, the next was that of the fusion image based on wavelet transform, and that of traditional pixel level fusion image is the worst. In addition, according to the fusion of images obtained from different types of sensors, it could be found, remote sensing image fusion effect was not only accordance with fusion algorithm, but also related to the types, phase and correlations between difference bands of fused image, and so on.In a word, the image fusion method combined with PCA, local algorithm and wavelet transform is an efficient image fusion method, fusion effect evaluation based on the classification accuracy also is a feasible method, which make fusion image data is directly facing the application.