Data mining is the process of extracting hidden and potential information in large datasets by artificial intelligence and other methods, which provides an effective way to obtain valuable knowledge from a large amount of information. Data mining is omnipresent in the process of solving point cloud registration task by deep learning. Extracting global features and estimating rigid body transformation are two key stages of corresponding-free point cloud registration. Mining abundant information hidden in two stages is one of the important tasks of point cloud registration. However, recently proposed methods are easy to ignore low-dimensional local features when extracting global features, resulting in the loss of numerous point cloud information, which makes the accuracy of solving transformation parameters in the subsequent rigid body transformation estimation stage unable to reach the expectation. Firstly, a features mining network based on multi-dimensional information fusion is devised, which fully excavates the high-dimensional global information and low-dimensional local information in point cloud, and effectively offsets the lack of local features in the global feature extraction stage of point cloud registration. Secondly, dual quaternion is utilized to estimate pose in the rigid body transformation estimation stage, which can represent rotation and translation simultaneously within a common framework and provide a compact and precise representation for pose estimation. Finally, extensive experiments on ModelNet40 dataset are conducted. The results illustrate that, compared with the existing corresponding-free point cloud registration methods, the proposed method can obtain higher accuracy, while being highly robust with respect to noise.