摘要
将小波引入SIFT算法,提出利用小波多分辨率分解降低图像中边缘、噪声等不良信息点比重方法。利用工程上常用的4类小波,分别对图像进行小波分解,将图像的低频成分建立尺度空间,基于局部特征提取图像特征点,引入统计学理论中的召回率和图像匹配中的准确率两个指标,研究不同小波对图像匹配中特征点提取的影响,寻求图像处理的最优小波类型。实验表明,bior4.4小波对图像的多分辨率分解后获得的特征点,可以在较低错配率的情况下获得较高的召回率。
Matching accuracy is an important target to determine matching algorithm's superior or inferior. Wavelet multi-resolution decomposition was used in SIFT algorithm to decrease the share of edge, noise pixels in the images. The images were decomposed with the four wavelets widely used, and then low-frequency component of the images were used for building scale space to extract image feature points based on local feature. Recall used in statistical theory and accuracy used in image matching were introduced to analyze the effect on extracting feature points at different wavelet domain, and then to find the best wavelet type used image processing. Results show that there is lower error probability while higher recalls by using bior4.4 on image matching.
出处
《西北大学学报(自然科学版)》
CAS
CSCD
北大核心
2015年第1期53-57,共5页
Journal of Northwest University(Natural Science Edition)
基金
国家自然科学基金资助项目(10872160)
陕西省自然科学基础研究计划重点基金资助项目(2011JZ012)