一种基于图像轮廓信息的新视点图像生成算法

对于摄像机任意运动采集的序列图像，提出了一种基于轮廓对应的新视点目标图像生成算法．在此算法中，首先检测并提取图像中的轮廓边缘，将边缘分解为直线段，根据直线段间的相对位置关系实现不同图像间直线段的匹配，从而实现整幅图像问匹配．采集的参考图像首先转化到与视点连线平行的平面（称做校正图像）．然后，通过对校正图像匹配点线性插值生成对应于新视点的校正图像；最后将该生成的校正图像变换至最终位置，从而实现新视点图像生成．实验证明，该算法可应用于任何种类的图像合成，所生成图像可以获得清晰的边缘，图像质量较高．

An image synthesis algorithm based on new viewpoint using contour edge correspondence information

A new algorithm of view synthesis from serial images taken by moving camera using contour edge information is proposed.The contour edges in images are first extracted and approximated to multiple line segments.Then,from the information of relative positions and orientations of these lines,the correspondence of line-pairs in different image can be detected.An index system is also proposed for digitally characterizing the configuration of a line-pair to augment the matching processing.The acquired images are first transformed into images(referred as reference rectifying images) in a plane paralleling with the viewpoint connection line.An intermediate rectifying image corresponding to new specified viewpoint is synthesized from the reference rectifying images.Then a post processing is implemented for transforming the intermediate rectifying image to final target image in correspondence with specified viewpoint and position.The experimental results show that the quality of synthesis image generated by the proposed approach is better than that by the conventional approach,and the approach can be applied to any kind of image synthesis.