Multi-Scale Analysis And Fusion For Space Motion Image Sequences

Space image has many features such as simple background, long distance target, high exposure and low exposure. Existing image fusion algorithms are difficult to solve the problems of image fusion under the space background. Therefore, based on the study of existing fusion algorithms, we completed the study of cross-scale analysis and fusion on space image sequences. The main work done includes the following aspects:(1) We proposed multi-source image fusion algorithm based on SCDPT transform and SSIM. Through SCDPT decomposition, we got the low frequency and high frequency coefficients of the image. For the low frequency coefficients, the fusion rule is based on structural similarity and regional energy, for high-frequency sub-band coefficients, the fusion rule is based on structural similarity and regional variance. Compared with different fusion algorithms, the proposed algorithm can extract image features better. Experimental results show that proposed algorithm has greatly improved the objective evaluation.(2) We proposed a fast image fusion algorithm based on salient region. In order to solve the spatial image fusion problems, for example, the important target of space image is always too big or too small and the background is always simple. Therefore, through salient region detection, we got the significant region and non-significant region of space image. For salient region, we adopted complex fusion strategy, and for the non-significant area we take simple fusion strategy. This greatly reduces the fusion time, and the important area will not be lost. Experimental results showed that this method can effectively carry out precise blend of important goals, and the fusion time is much less than UDCT, NSCT. (3) We proposed a fast image fusion algorithm based on background separation. The background of spatial image is always simple and large, in order to build a real-time image fusion system for space images, background fusion should not be the same with foreground. Therefore, we separated the original image into foreground and background information with the adaptive threshold method. For the background, we took simple fusion rules and for the foreground, we adopted complex fusion rules. From the experiment of the standard image, our algorithm can guarantee the quality of the image fusion, and it takes less time.(4) We finished the design and implementation of multi-source image fusion toolkit. The tool is able to meet most of image fusion capabilities, and has high scalable, high performance, and better user interaction.