基于塔形分解的局部熵差图像配准

以图像局部熵差为匹配准则,确定关键点的匹配位置。逐点计算图像局部熵,将图像局部熵序列进行塔式分解。采用金字塔式的数据结构,通过从低分辨率图像开始模板匹配,找出粗匹配点,逐步找到原始图像(即高分辨率图像)的精确匹配点,大大减少了计算量。该算法具有良好的抗噪声能力和抗几何失真能力。实验结果表明,当实时图相对于参考图旋转不超过5°时,正确匹配达到76%以上;当椒盐噪声强度不超过5%时,正确匹配达到78%以上;当零均值高斯白噪声方差不超过0.02时,正确匹配达到70%以上。

Image matching based on local entropy and pyramid transform

A method to determine image matching positions of key points based on criteria of local entropy difference is proposed in the paper. Local entropy of each point of image is computed point by point, and the sequences of local entropy are analyzed by means of pyramid transform, i.e. through pyramidal data structure to find crude matching point of low resolution image, and determine accurate matching point of original image (high resolution image) finally. The method has raised the computation effect, featuring good anti-noise properties and anti-distortion of geometry. Experiment results show that its matching accuracy is up to 78% when intensity of salt & pepper noise is under 5%.