摘要
为解决航拍地图拼接计算量大、内容复杂多变的问题,提出使用超像素尺度不变特征变换(SIFT)特征进行地图拼接。首先利用近邻传播(AP)算法改进简单线性迭代聚类(SLIC)算法实现超像素自适应分割。然后在引导信息下对少数超像素提取SIFT特征,进行SIFT特征匹配。最后在线生成监督机制,在随机抽样一致性(RANSAC)算法配合下,得到少而精的匹配点,计算变换矩阵融合图像,实现图像拼接。实验结果显示:对于较高分辨率的图像,该算法与单纯SIFT拼接算法相比,在计算时间和拼接质量上都显示出较大的优势。
In order to address dynamically changing contents and large computation in aerial image mosaic,propose a region-matching algorithm based on scale invariant feature transform（ SIFT） features of superpixels.Firstly,improve a simple linear iterative clustering（ SLIC） by affinity propagation（ AP）,and the method is used to accomplish adaptive superpixel segmentation; Next,a few superpixels are described with SIFT features to acquire coarse point matching under the guidance. Finally,supervision is obtained online to select more superpixels to be matched and random sample consensus（ RANSAC） method is constructed so that aerial image mosaic is accomplished precisely. Experimental results demonstrate that,compared with SIFT mosaic algorithm in high resolution,the method greatly reduces computation time and improves quality of aerial image mosaic.
出处
《传感器与微系统》
CSCD
2016年第12期17-19,23,共4页
Transducer and Microsystem Technologies
基金
公益性行业(气象)科研项目(GYHY200706004)
天津市自然科学基金资助项目(09JCYBJC07500)
