基于局部特征射影变换的组织切片图像非刚性配准

针对组织切片图像配准问题,研究一种以局部特征射影变换配准方法为基础的全局非刚性配准方法以获得更好的匹配效果。首先采用空域增强与频域增强结合的方法对图像进行预处理,随后应用匹配滤波突出组织四周的轮廓特征,之后提取处理后的图像的SIFT特征并进行初步的匹配,根据匹配的特征点的坐标通过RANSAC方法计算两幅图像之间的射影变换矩阵参数并进一步剔除离群点,该矩阵可用于全局配准,之后同样根据RANSAC方法提取出的特征点采用K近邻方法对图像进行区域划分,针对不同局部区域单独求解射影变换矩阵,并与全局配准矩阵进行比较筛选得到组织切片全局非刚性配准模型。实验结果表明对于不同的组织切片图像,局部射影变换配准的方法基本都可以提高配准的准确率,但对于局部变形较小的图像对,局部射影配准对配准精度的提高有限。

Non-Rigid Registration of Tissue Section Images Based on Local Feature Projective Transformation

Aiming at the problem of tissue section images registration, a global non-rigid registration method based on local feature projective transformation was studied to obtain better matching effect. Firstly, the spatial domain enhancement and frequency domain enhancement were combined to preprocess the image,then matched filter was applied to highlight the contour features around the tissue. After that, the SIFT features of the processed image were extracted and preliminarily matched. According to the coordinates of the matched feature points, the projective transformation matrix parameters between the two images were calculated by RANSAC method, and the outliers were further removed. The matrix can be used for global registration, and then according to the feature points extracted by RANSAC method, the k-nearest neighbor method was used for image region divided, and projective registration matrix was solved for different local regions separately. Compared with the global registration matrix, the global non-rigid registration model of tissue section images was obtained. The experimental results show that for different tissue section images, the local projective transformation registration method can basically improve the registration accuracy, but for the image pairs with small local deformation, the improvement of the registration accuracy of the local projective registration is limited.