作者:崔领袖,张奇志,周亚丽
单位:北京信息科技大学 自动化学院,北京 100192
中图分类号:TP391.4
文献标识码:A
文章编号:1006-883X(2018)04-0023-06
收稿日期:2018-03-25
摘要:提出了一种基于哈尔小波分解变换和高斯尺度空间的图像特征点匹配算法。首先利用哈尔小波变换对基础图像进行3层行列分解,然后利用高斯函数卷积核对这些分解图像,进行尺度变换。提出了一个小波高斯金字塔塔林的概念,即对通过小波变换产生的多张不同分辨率的基础图像分别进行高斯尺度变换进而产生一个个独立的高斯金字塔,进而产生独立的高斯差分金字塔林,完成特征点检测。再引进规范化强对比度描述子对特征点进行描述。结果表明:Haar-Gaussia&NICD算法的效果和SIFT算法相当,特征点数量优于SIFT算法,在局部特征匹配方面要更有优势;而且和NICD描述子搭配使用,在运行速度方面要比SIFT算法更快。
关键词:哈尔小波变换;尺度空间;金字塔塔林;特征点检测;特征匹配
An Algorithm for Image Feature Point Matching—Haar-Gaussian&NICD Algorithm
CUI Ling-xiu, ZHANG Qi-zhi, ZHOU Ya-li
School of Automation, Beijing Information Science and Technology University, Beijing 100192, China
Abstract: An algorithm based on Haar wavelet transform and Gaussian scale-space is proposed for image feature point matching. First the basic image is carried out three-layer row-column decomposition with Haar wavelet transform, and then scale-space transformation with Gaussian kernel. A conception of Haar wavelet-Gaussian Pyramid forest is proposed, which means the basic images with different resolutions produced by Haar wavelet transform are respectively carried out Gaussian scale-space transformation to generate a series of independent Gaussian pyramids, and then generate independent Gaussian difference pyramid forest to complete detection of feature points. After that the Normalized Intensity Contrast Descriptor(NICD) is introduced to describe the feature points. The experimental results show that the Haar-Gaussia&NICD algorithm has similar performance to SIFT, more feature points than SIFT, which is more advantageous in local feature point matching, and is faster than SIFT in terms of operating speed with the introduction of NICD.
Key words: Haar wavelet transform; scale-space; Gaussian Pyramid forest; feature points detection; feature matching
阅读全文
备注:2018年 第24卷 第04期