基于改进SIFT和位置方向距离函数的胃肠道图像匹配算法

Gastrointestinal Tract Image Matching Algorithm Based on Improved SIFT and Position Direction Distance Matching Function

针对传统图像匹配算法在胃肠道图像匹配中受到胃肠道图像颜色差异大、纹理复杂和灰度差异等因素干扰,使得匹配结果存在正确匹配对较少、匹配时间长等问题,提出了一种基于改进SIFT和位置方向距离函数的图像匹配算法。首先,利用新的梯度定义消除胃肠道图像对之间的非线性灰度差异,重新进行尺度空间构建;再采用对数极坐标建立描述子,建立适合胃肠道图像特征的描述子。在胃肠道图像匹配阶段,构建了位置方向距离匹配函数进行特征匹配,利用随机抽样一致性(RANSAC)和模式搜索方法进行误匹配删除后完成匹配。利用胶囊机器人获取的胃肠道图像作为实验数据,与SIFT、SURF和ORB等算法进行对比实验。实验结果表明,提出的算法对于胃肠道图像不仅在匹配精度和匹配时间上有较大的提高,还具备较强的鲁棒性。

The traditional image matching algorithm is interfered by factors such as large color difference,complex texture and gray difference of gastrointestinal image in gastrointestinal image matching,which makes the matching results have problems such as less correct matching pairs and long matching time.To solve these problems,an image matching algorithm based on improved SIFT and position direction distance function was proposed.First,the new gradient definition was used to eliminate the nonlinear gray difference between gastrointestinal image pairs and reconstruct the scale space.In addition,descriptors suitable for gastrointestinal tract image features were established by using logarithmic polar coordinates.In the phase of gastrointestinal tract image matching,feature matching based on location and direction distance matching function was constructed,and random sampling consistency(RANSAC)and pattern search method were used to delete mismatches and complete matching.The gastrointestinal tract images obtained by the capsule robot were used as experimental data and compared with SIFT,SURF and ORB algorithms.Experimental results show that the proposed algorithm not only improves the matching accuracy and matching time,but also has strong robustness in gastrointestinal tract images.