| Aimed at the fusion of multisensor images with different resolutions in the fusion field, and translation, rotation, scaling and skewing of objects and damaged boundaries in the recognition filed, the following research is done.1.Aimed at the fusion of images with different resolutions in the fusion field, a multiresolution image fusion algorithm which combines block modeling and probabilistic model is proposed. First, every vector-model of the corresponding blocks from different source images was given by using block-stack modeling method; second, modeled every block vector-model by probabilistic model, the maximum a posterior estimate of every block vector-model was obtained according to Bayesian rule; next, the fusion result of every block was acquired after the parameters of block vector-model was calculated by E-M algorithm; finally, all of block estimates was constructed to get the fusion result of different multiresolution images. The computer simulation shows that our algorithm is effective to images with different resolutions, compared to the traditional method just by using the wavelet transform.2.Aimed at translation, rotation, scaling and skewing of objects and damaged boundaries in the recognition filed, a new method for object recognition based on affine statistical invariants was proposed. First, the affine statistical invariants was derived based on the affine arc length parameter which is invariant; second, based on the comparison between different edge detecting operators, a method combining the Canny algorithm with the boundary tracking algorithm is adopted to obtain the ordered and single pixel set, and then the obtained edge set was applied to calculate the affine statistical invariants; finally, the statistical sort method based on distance function was used to recognize the desired objects. The experimental results show that the proposed method is effective, and to some extent, can be used to overcome the effect of the bad boundary to recognition result. |
PDF Full Text Request