Multi-source Image Sequence Information Fusion Based On Feature And Multi-scale Transformation

Multi-source image sequence is a kind of multiple images provided by different image sensor in the same scene, which plays an important role in various fields of real life. Due to the difference among different image sensor, such as the operational principle, installation position and focus parameter, the picture which takes may also exist many differences such as the relative position shift, different focus, image sequence information changes. Therefore, we need to study effective algorithms to do registration and fusion on these multi-source motion images, comprehensively utilize the redundant information provided by multiple sensors. By this way, the fusion images contain more detail information. The main contents and achievements of this thesis are as follows:(1) A novel multi-focus image sequence fusion algorithm based on region feature is proposed. The average image is segmented based on the region feature. And the region coordinates of the segmented region are mapped to the coefficients of source images obtained by discrete wavelet frame transform, which make full use of the region features. Compared with the image fusion algorithm based on Laplace, Contourlet transform and translation invariant discrete wavelet transform, experimental results show that the proposed algorithm achieves a better entropy index by 3 .3%,3.5%,spatial frequency by 5.3% and 2.5%,QABF by 2.2% and 1.5%, and mutual information by 11.5% and 8.6%.(2) An image sequence fusion algorithm based on multi-scale transformation is proposed. In the process of image fusion, DSIFT feature descriptors are used to compute the active level of the image. The initial fusion rule is generated by sliding window technique. The normalized DSIFT feature descriptor and spatial frequency are used as metrics to generate the final fusion rules efficiently. The experimental results show that the fusion algorithm based on multi-scale transform achieves a good performance in the experiment. Compared with the traditional image sequence fusion algorithms, the average gradient is improved by 15.9%,19.4%, 22% and 9% on average in four groups of experimental data, the edge strength is increased by 14.6%, 13.2%, 14.7% and 10.8%, the spatial frequency is increased by 13.6%, 12.0%, 4.5% and 8.1%, the mutual information is increased by 20.2%, 19.0%, 10.0% and 27.9%.(3) An image sequence fusion algorithm based on texture features and gradient features is proposed. In the process of image decomposition,texture features and gradient features are included in the transformed domain, providing more comprehensive measurement of information. The experimental results show that, compared with existing image sequence fusion algorithms, the fusion algorithm proposed in this thesis has improved the entropy by 0.7% and 1%, the spatial frequency by 15% and 14.4%, the average gradient by 18.2% and 17.4% and the mutual information by 17.5% and 31%.(4) A multi-source image sequence information fusion verification system based on multi-scale transform is developed to verify the effectiveness of the proposed algorithms. It also includes some classic image sequence fusion algorithms as an optical contrast algorithm. Besides,an evaluation module is implemented to evaluate the performance of image sequence fusion algorithm.