研究目的

Investigating the therapeutic effects of a specific herbal medicine on a particular disease.

研究成果

An automatic image alignment using PCA was proposed in this paper. PCA was used to assess the orientation of the digits, fingerprints, and T1-weighted brain MRIs automatically, thus streamlining the subsequent tasks in optical character recognition, automatic fingerprint matching, and registration in medical images of different modalities. PCA aligned the data in their principal spread. However, the existing PCA methods demonstrated problems of 180? rotation in its principal axis due to the random nature of its principal components. The problem was solved by an assignment-based algorithm. The proposed algorithm functions efficiently for data having a maximum variance towards its true side. In the developed algorithm, the rotation angle range [-90?, 90?] was considered from any standard orientation. If the rotation angle was beyond the specified range, the standard orientation was 180? to the original orientation. The proposed technique is reasonably insensitive to outliers as it exploits all the coordinate points of the ROI in forming the covariance matrix for PCA. The algorithm is efficient and can be used as a preprocessor in the presented applications. Experimental results on real datasets further corroborated the effectiveness and robustness of the proposed algorithm.

研究不足

The presented algorithm works well for images in which the principal axis is well distinguished. However, as PCA based alignment methods do, the alignment becomes less robust for the images which have a strong symmetry. In this case, it could be still used as an initial guess of rotation angle for classes of images. Another limitation is that the overall performance of the algorithm depends on the success or failure of ROI extraction. If there exists severe noise in the images, the sophisticated ROI extraction should be employed before applying the proposed algorithm.