摘要
序列图像配准是光学影像超分辨率重建应用中的关键技术,针对如何提高序列图像超分辨率重建过程中配准的速度和精度问题,提出了一种基于小波分解和改进的多约束准则相结合的配准方法。首先对参考帧和待配准帧进行小波分解,生成小波图像金字塔以缩小搜索空间。其次,从金字塔最高层开始利用由局部灰度熵准则、灰度相似性准则和简化欧式距离比例不变准则构成的改进多约束准则提取同名特征点对,利用最小二乘法计算初始配准参数,然后逐层向下,对配准参数进行修正,实现由粗到精的配准。最后,利用模拟生成图像序列与实际获取图像序列进行测试。实验结果表明该方法在参数获取达到较高精度的情况下能有效提高配准速度,具有较好的效果。
Image sequence registration is a key technique for super-resolution reconstruction.To improve the accuracy and speed of image registration,a new method was proposed to register consecutive frames based on wavelet transformation and improved multi-restriction criterion.At first,the wavelet image pyramids of the reference frame and sensed frame are generated to narrow the search space.The feature points are found using Harris detector.Then using the improved multi-restriction criterion,the matching feature points are extracted from the highest level of the image pyramids.Least squares technique is employed to calculate the initial registration parameters.Then the coarse-to-fine hierarchical strategy is applied.The estimates of the mapping function parameters are gradually improved at the following level of the pyramids.Finally,simulated images and actual images were used to carry out test.Experimental results on the image sequences demonstrate that the presented method can quickly obtain registration parameters with high accuracy.
出处
《仪器仪表学报》
EI
CAS
CSCD
北大核心
2011年第10期2261-2266,共6页
Chinese Journal of Scientific Instrument
关键词
图像配准
小波多分辨
局部灰度熵准则
灰度相似性准则
简化欧式距离比例不变准则
image registration
wavelet multi-resolution
local grayscale entropy criteria
gray similarity criteria
simplified unchanged criteria of Euclidean distance