Research Of Multi-sensor Image Fusion Algorithms Based On Multi-resolution Analysis

Image fusion is one of image processing techniques. It fuses two or more images whichhave complementary informations into one single image which is more substantial, moreperfect and reliable. There will be less redundancy and less fuzzy in the fused image. Thefused image has better complementarity and intelligibility. It is more suitable for humanvisual system, computer detection, computer recognition, and so on. Nowadays image fusionhas been widely used for remote sensing image processing, computer vision, intelligent robots,military monitoring, medical scanning and imaging.This paper mainly researches multi-sensor image fusion techniques based onmulti-resolution analysis. We utilize multi-resolution analysis tools such as Dual-TreeComplex Wavelet Transform, the second generation Curvelet Transform andNon-subsampled Contourlet Transform to research multi-sensor image fusion techniquesdeeply. A lot of new algorithms have been proposed combine with an amount of simulationexperiments.Firstly, the classification of multi-sensor image fusion is kept as the pointcut to analyzethe image fusion techniques based on multi-resolution analysis. Then distribution laws ofvarious mainstream multi-resolution analysis tools coefficients are compared. Imagingproperties of three types of sensors are introduced, they are multi-focus image sensors,infrared image sensors and remote sensing image sensors. Evaluation indicators andevaluation basises of image fusion are concluded.Secondly, a gray multi-focus image fusion algorithm based on fuzzy theory classificationand Dual-Tree Complex Wavelet Transform is proposed in accordance with fusion of graymulti-focus images. Fuzzy theory is introduced to remove uncertainty when choosingsub-band coefficients in smooth regions after Dual-Tree Complex Wavelet Transform. Fuzzymembership is used to measure the mapping relation between different feature types ofimages and corresponding pixels gray values. The uncertainty probably existes in fusionprocess is removed efficiently. Efficient fusion and detail enhancement of images are realized.On this basis, a color multi-focus image fusion algorithm based on IHS color space andDual-Tree Complex Wavelet Transform is proposed in accordance with fusion of colormulti-focus images. This method could solve the problem of color distortion in traditionalalgorithms. Simulation results verify the effectiveness and superiority of proposed algorithms. Thirdly, according to the fusion of infrared images and visible images, combines withthe second generation Curvelet Transform, a infrared and visible image fusion algorithmbased on salience measure and regional matching rules is proposed. According to thealgorithm, the salience measure operator is defined combining with characteristics of theinfrared image, the visible image and the human visual system. It represents informationquality of pixels and guides the selection of fusion coefficients after the second generationCurvelet Transform. Then characteristic informations of infrared image’s hot targets and thatof visible image’s abundant background are extracted fully to obtain better fusionperformance. On this basis, a new infrared and color visible image fusion algorithm based onHSV color space and the second generation Curvelet Transform is proposed, which makes thefused image retain natural color information from the visible image as more as possible.Finally, a remote sensing image fusion algorithm based on NSCT and PCA is proposedin accordance with fusion of remote sensing images. Multi-scale characteristic of NSCT andthe characteristic which PCA transform can reduce dimensions are used sufficiently. Thenfused image could keep spectrum information from the source image as more as possible. Wecan remove the drawback that space quality was reduced in traditional image fusionalgorithms based on multi-resolution analysis. According to the problem of spectrumdistortion from injection of the panchromatic image into the principal component ofmulti-spectral image, the conception of high pass is introduced to sharp the panchromaticimage. Then major spectrum information can be kept in the fused image. In the fusion rules,in choosing the low-frequency coefficient, this paper proposes a fusion strategy based onwindow combines with local variance. In choosing the high-frequency coefficients, a methodbased on regional linear relativity measure is proposed. The experimental results show thatthe algorithms proposed in this paper could improve space quality and reduce spectrumdistortion of fused image.From all this, this paper researches image fusion techniques based on multi-resolutionanalysis. Then kinds of algorithms are designed to remove the deficiencies in the currentimage fusion. Simulation experiments show that the algorithms proposed in this paper obtainbetter fusion results.