基于融合特征和FAST-SURF的蚕茧疵点图像配准算法

Registration Algorithm for Cocoon Defect Image Based on Fusion Features and FAST-SURF

针对获取的蚕茧完整表面图像疵点横跨两张不同角度图像的问题,文中提出一种基于融合特征与FAST-SURF的蚕茧疵点图像配准算法。首先通过HSI彩色空间获取蚕茧疵点图像,提取疵点图像的Canny边缘特征和LBP纹理特征,分别进行归一化后构成融合特征;其次,在蚕茧疵点融合特征灰度图上利用FAST算法快速检测特征点,并为每个特征点构建64维SURF特征描述符;最后,结合坐标距离筛除与RANSAC算法获得最终匹配对,实现蚕茧疵点图像配准。结果表明,将蚕茧疵点图像的融合特征与FAST-SURF算法相结合得到的匹配正确率均值达89.76%,配准总时间均值为0.46 s。建立的算法满足蚕茧疵点图像快速配准的精度与实时性要求。

Aiming at the problem that the defect spans two images from different angles when acquiring the complete surface image of the cocoon,a registration algorithm for cocoon defect image based on fusion features and FAST-SURF is proposed in this paper.Firstly,the cocoon defect image is obtained through HSI color space,and the Canny edge feature and LBP texture feature of the defect image are extracted and normalized to form the fusion feature.Secondly,the FAST algorithm is used to quickly detect feature points on the grayscale image of cocoon defect fusion feature,and a 64-dimensional SURF feature descriptor is constructed for each feature point.Finally,the final matching pair is obtained by combining the coordinate distance screening and the RANSAC algorithm to realize the image registration of the cocoon defect.The results show that the average matching accuracy obtained by combining the fusion feature of the cocoon defect image with the FAST-SURF algorithm is 89.76%,and the average total registration time is 0.46 s.The algorithm proposed in this paper satisfies the accuracy and real-time requirements of fast registration of cocoon defect images.