Geometric Global Registration of Parallax Images Based on Wavelet Transform

Image registration is an important task of image processing. In parallax image mosaic, the geometric registration is a global registration minimized residual in translation model. In order to realize quick and robust geometric registration of parallax image, a coarse-to-fine global registration strategy using the coarse images weighted with detail image is proposed. Taking the advantage of multi-resolution characteristic of Wavelet, the original image is pyramid decomposed by Wavelet transformation. The registration result of the upper layer can be the initial value of the next layer in the global registration. The better initial value by this way is applied in the registration of coarse image and vertical detail image in the layer next to the bottom, and the correlative registration curve can be got. The extremums can be got from the correlative registration curve which weights the weight vector by means of the correlative registration curve of the vertical detail image in the layer next to the bottom. Then the optimal registration can be achieved. Experiments indicated that this method has high accuracy, reliability and real-time property.