SIFT特征算法改进研究  

Research on the Improvement of SIFT Feature Algorithm

作　　者：代玉强 杨絮[1] 熊昊 陈小远 莫苏新 DAI Yu-qiang;YANG Xu;XIONG Hao;CHEN Xiao-yuan;MO Su-xin(School of Electronic Information Engineering,Changchun University of Science and Technology,Changchun 130022)

机构地区：[1]长春理工大学电子信息工程学院,长春130022

出　　处：《现代计算机（中旬刊）》2018年第6期55-59,共5页Modern Computer

摘　　要：SIFT算法在目标识别算法上有广泛的应用,但经典的SIFT算法忽略图像的颜色信息,并且特征描述向量维数太高,计算复杂。针对这些问题,融入颜色不变量理论,先求出彩色图像的颜色不变量,以此作为输入图像提取特征点,提高算法对于颜色信息的敏感度。然后将特征描述区域设置成圆形,降低特征描述子维度。SIFT algorithm has a wide range of applications in image recognition,but the classic SIFT algorithm ignores the color information of the image,and feature description vector has high dimensions,computational complexity.To deal with these problems,at first,incorporates color invariant theory and finds color invariants of color images,as the input image to extract feature points,improves the sensitivity of the algorithm to color information,and uses the characteristics of the circle to set the feature description area,reduces the feature descriptor sub-dimension.

关 键 词：SIFT算法 颜色不变量 目标识别 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]