基于纹理分类的多阈值SIFT图像拼接算法  

Multi-Threshold SIFT Image Stitching Algorithm Based on Texture Classfication

作　　者：钟岷哲 唐泽恬 王昱皓 杨晨 ZHONG Min-zhe;TANG Ze-tian;WANG Yu-hao;YANG Chen(College of Big Data and Information Engineering,Guizhou University,Guiyang Guizhou 550025,China)

机构地区：[1]贵州大学大数据与信息工程学院,贵州贵阳550025

出　　处：《计算机仿真》2022年第10期364-368,共5页Computer Simulation

基　　金：国家自然科学基金(61604046);国家自然科学基金(62065003);贵州省科技计划项目(黔科合平台人才[2017]5788号,[2018]5781号);半导体功率器件可靠性教育部工程研究中心开放基金(黔科合平台人才20176103号)。

摘　　要：针对传统SIFT算法的单一阈值导致的特征点空间分布不均匀的问题,提出了一种纹理分类的多阈值SIFT图像拼接算法。算法对图像的纹理复杂度进行了分析和分类,并增加纹理分类后的弱纹理区和一般纹理区的候选特征点,减少强纹理区域的候选特征点。通过信息熵设置弱纹理区和一般纹理区的阈值,以均衡特征点的数量。在以上基础上进行特征点匹配和图像融合完成图像拼接。相比传统的方法,实验结果表明,纹理分类的多阈值方法检测的特征点在空间分布上更加均匀,更有利于拼接。基于纹理分类的多阈值SIFT图像拼接算法可有效地提高拼接质量,在对图像质量有较高要求的场景中有潜在的运用价值。A multi-threshold SIFT image mosaic algorithm for texture classification is proposed to solve the problem of uneven spatial distribution of feature points caused by a single threshold of the traditional SIFT algorithm. The texture complexity of the image was analyzed and classified in the algorithm, then candidate feature points were enhanced in the weak and the general texture area, while reduced in the strong texture area. In order to balance the number of feature points, the threshold was set according to the information entropy for the weak and general texture area. Subsequently, feature point matching and image mosaic were carried out. Compared with the traditional method, experimental results show that the spatial distribution of feature points detected by the multi threshold method based on texture classification is more evenly, which is beneficial for image mosaic. Therefore, the SITF image mosaic algorithm based on multi threshold texture classification can effectively improve the mosaic quality, which has potential application value in scenes with high requirements for image quality.

关 键 词：纹理分类 多阈值 图像拼接 信息熵 

分 类 号：TP391.9[自动化与计算机技术—计算机应用技术]