A Technique For Color Image Mosaic For Cylindrical Panorama

IBR (image-based rendering) is a new attractive approach in computer vision and computer graphics, which represents the scene with a sequence of pre-acquired images rather than geometric primitives and synthesizes novel views by properly combining these images.First, we identify the homonymous points using normalized product correlation to revise the geometry distortion. In this method, the linear combination of correlation's three features is considered to improve the matching probability of homonymous points. And then, the eight-parameter 2D projection model is optimized by GA, including adjusting genetic operators to eliminate the local convergence and premature convergence problem. Image fusion based on wavelet transform is presented. By decomposing the images and mask signals with wavelet transform, wavelet coefficients and approximation coefficients at different scales are obtained. With the corresponding coefficients of normalized-mask signals as weighting functions, coefficients are combined in different resolution levels to reduce the effect of the seam. And the fused image can be obtained by using the inverse wavelet transform for combined wavelet coefficients and approximation coefficients.The merits of our algorithm include following: (l)The camera needn't be calibrated,and imposes no strict limit on camera rotation angle between conjoint images. (2)The registration image accords with projection model to ensure that the registration image is real. (3)The results of experiments using image fusion based on wavelet transform show that the combined image is smooth and almost seamless. This image stitching technology can be used in a wide range and it has an ideal stitching effect.