李 丽, 郭双双, 梅树立, 张楠楠. 基于特征点提取匹配的蝗虫切片图像的拼接和修复方法[J]. 农业工程学报, 2015, 31(7): 157-165. DOI: doi:10.3969/j.issn.1002-6819.2015.07.023
    引用本文: 李 丽, 郭双双, 梅树立, 张楠楠. 基于特征点提取匹配的蝗虫切片图像的拼接和修复方法[J]. 农业工程学报, 2015, 31(7): 157-165. DOI: doi:10.3969/j.issn.1002-6819.2015.07.023
    Li Li, Guo Shuangshuang, Mei Shuli, Zhang Nannan. Mosaic and repair method of locust slices based on feature extraction and matching[J]. Transactions of the Chinese Society of Agricultural Engineering (Transactions of the CSAE), 2015, 31(7): 157-165. DOI: doi:10.3969/j.issn.1002-6819.2015.07.023
    Citation: Li Li, Guo Shuangshuang, Mei Shuli, Zhang Nannan. Mosaic and repair method of locust slices based on feature extraction and matching[J]. Transactions of the Chinese Society of Agricultural Engineering (Transactions of the CSAE), 2015, 31(7): 157-165. DOI: doi:10.3969/j.issn.1002-6819.2015.07.023

    基于特征点提取匹配的蝗虫切片图像的拼接和修复方法

    Mosaic and repair method of locust slices based on feature extraction and matching

    • 摘要: 由于环境和切片本身特性的影响,试验中获取的蝗虫切片总是不完整或者带有褶皱的。针对这一问题,提出了一种基于图像匹配的蝗虫切片拼接和修复方法,以序列切片中缺损切片的邻近切片图作为参考对象,对缺损切片图像进行拼接和修复。首先对切片进行小波降噪,降低噪声对匹配的影响;然后用尺度不变特征变换(scale-invariant feature transform, SIFT)算法和快速鲁棒特征(speed-up robust features, SURF)算法获取切片特征点并生成特征向量,完成切片的初始特征点匹配;随后利用随机采样一致(random sample consensus, RANSAC)算法进行匹配矫正,剔除匹配点对中的误匹配,并利用最小二乘法求解出单应性矩阵;最后用正确的匹配点对和单应性矩阵完成蝗虫切片的拼接,利用求得的空间映射模型找到褶皱部分相应的匹配域,完成对缺损部分的修复。试验表明:提出的切片拼接和修复方法的所提取的特征点的正确匹配率能够达到72.2%,并且运行速度以及匹配效果都能在一定程度上满足切片修复的要求,为后面进行蝗虫切片褶皱打开以及蝗虫体的三维重建提供了技术参考。

       

      Abstract: Abstract: Because of the influence of the environment and the characteristics of the locust slices, locust slices obtained from experiments are always incomplete or folded. To match and reconstruct this folded area, the slice in the sequences adjacent to the slice which needs to be repaired is used as the reference image, and in this way, the slice is repaired successfully. The image mosaic and repair algorithm used in the article has the following steps: image preprocessing, feature extraction, feature matching, coordinate transformation and image mosaic and repair. Firstly, as there is some noise in the slice image that can influence the repairing effect, we use wavelet denoising on images to reduce the effect of noise on feature extraction before searching the feature points; secondly, as there are no related materials about how to match or repair locust slices, in order to find more suitable feature extraction methods for locust slices, we compare the scale-invariant feature transform (SIFT) with speed-up robust features (SURF) algorithm which are already widely used in computer vision and image registration, and then generate the image feature vectors to complete the feature extraction from the initial slice image, record the scale, position and direction of the feature points at the same time, and finish the preliminary matching between the two slices; thirdly, as the preliminary matching can not completely remove the error matching points, random sample consensus (RANSAC) algorithm is used to correct the matching errors and eliminate the error matching points by repeatedly choosing a group of random subsets of the data, finally leaving the correct matching points; fourthly, the homography matrix is calculated by using the least square method based on all the correct matching points, which makes full use of existing points to calculate and therefore makes the results have more universal adaptability; finally, the locust slices mosaic is finished using the correct matching points and the homography matrix, the corresponding matching block for the folded part is found out using space mapping model and the slice repair is completed. The experiment adopts two groups of positive images of locust slice obtained from the experimental environment under the microscope. What's more, statistical results of contrast experiment with SIFT and SURF algorithm are obtained, each of which shows the running time, the number of matching points and matching rate. The experimental results show that: the matching rates reach 72.2% and 34.0% using SIFT algorithm and SURF algorithm, respectively, which means that SIFT algorithm can meet the requirements of image mosaic more precisely compared with SURF algorithm. As the matching rate doesn't perform as well as expected, the error sources such as the change of content or the effect from the fold are briefly analyzed. The registration algorithm provides the reference for the later work of the folded area's opening on the slice image and the three-dimensional reconstruction of locust.

       

    /

    返回文章
    返回