Research On Pixel-level Image Fusion And Its Key Technologies

The image fusion means to integrate and synthesize the information of the source images for the same scene acquired with the same or different kinds of image sensors, and generate a single image which contains a more accurate description of the scene than any of the individual source images. The image fusion can be divided into three fusion level, namely: pixel, feature and decision levels. Among unsolved key problems of the pixel-level image fusion, we mainly study three key technologies, namely: the image denoising technology in the pixel-level image fusion preprocessing, the multi-focus image fusion technology, and the panchromatic and multispectral remote sensing image fusion technology. The main results are as follows:1. An image denoising scheme based on human visual system is proposed. This method combines the pixel classification with the wavelet transform and denoises the different image areas with the different thresholds respectively. This approach reduces the image noise effectively and keeps the image details well.2. An image denoising algorithm suitable for image compression and a wavelet coefficient verification method are proposed. This denoising method exploits the intrascale wavelet coefficients' correlation to denoise the image. This approach can be combined with the succedent image compression operation effectively.3. A local gradient information based multiresolution image fusion algorithm and its modified method are proposed. The modified method uses adaptive weighted addition to obtain the fused scale coefficients according to the corresponding scale coefficients of the different source images. The fusion results of the proposed methods are superior to those of conventional fusion schemes.4. An image fusion algorithm based on the discrete cosine transform (DCT) and a new image fusion scheme using the wavelet transform and DCT are proposed. The former method with the low computational cost is more suitable for the real-time processing, while the latter one can improve the fusion quality effectively.5. An adaptive multi-focus image fusion algorithm based on the support vector machine (SVM) and image block segment is proposed. The original images are fused adaptively with different block sizes according to the positions of the original image blocks with the proposed scheme. This method can improve the fusion quality effectively.6. An image fusion algorithm based on multi-resolution and image block segment is proposed. The proposed method can be combined with the existing multi-resolution based multi-focus image fusion algorithms and improves the fusion results of these methods.7. A multispectral and panchromatic remote sensing image fusion algorithm using discrete cosine transform and Intensity-hue-saturation (IHS) transform and its modified approach are proposed. The proposed approaches can be performed in the DCT compression domain directly and are suitable for fast image fusion in the compression domain. The idea of the suggested methods combined with the traditional IHS-based image fusion scheme can be employed to improve the spatial quality of the fused image and keep the spectral characteristics of the green vegetation areas with the low computational complexity.8. A remote sensing image fusion algorithm with high spatial quality based on IHS transform and the decimated wavelet transform is proposed. The computational cost of the proposed method is close to that of the conventional decimated wavelet transform based fusion approach. The fusion results of the proposed fusion scheme are similar and even superior to those of the traditional undecimated wavelet transform based fusion algorithm.Computer simulation and its performance analysis are carried on with all techniques discussed above. All the research works in this dissertation have important values in the theory and application on image denoising and image fusion fields.