Search

Nonuniform illumination is one of the main issues that hinder the accuracy of face recognition because it makes the intra-person variation more complicated. To minimize the intra-person differences caused by varying illuminations, this paper presents a normalization method based on Convolutional Auto-encoder (CAE). The CAE is employed to map the face images under various illumination conditions to a normalized one, generating preliminary results with blurry and insufficient facial details, which are tricky for recognition. To recover these details, a restoration method based on re-blurring strategy and frequency analysis is proposed to preserve the facial features lying in high-frequency components based on discrete cosine transform (DCT). Therefore, in our method, these components are extracted and re-introduced into the outputs of CAE to enhance the fidelity of outputs. Thus, the facial details are preserved to the largest degree and the following works such as recognition tasks are benefited. Experiments conducted on the AR, extended Yale B, and CAS-PEAL database demonstrate the effectiveness of our method.

We propose a registration algorithm based on neighborhood similarity for 3D point clouds collected by optical measurement and without prior information. The algorithm first applies the improved minimum spanning tree (Prim algorithm) to classify the point cloud in order to obtain the topology information of the data. Specifically, vectors among root nodes and child nodes are processed, and the points on nodes are classified into different levels according to their scanning angle to simplify data and preserve the most representative points. Then, through the perspective conversion between 2D and 3D and according to the corresponding point set obtained by previous classification, the fast normalized cross-correlation (a 2D matching criterion) is applied to determine the relationship between initial characteristic points. Finally, distance constraints remove the errors between point pairs and allow calculating the registration parameters. Experimental results show that the algorithm has high registration accuracy and is suitable for point cloud data obtained by laser and structured light acquisition.

A novel calibration method based on polar coordinate is proposed. The world coordinates are expressed in the form of polar coordinates, which are converted to world coordinates in the calibration process. In the beginning, the calibration points are obtained in polar coordinates. By transformation between polar coordinates and rectangular coordinates, the points turn into form of rectangular coordinates. Then, the points are matched with the corresponding image coordinates. At last, the parameters are obtained by objective function optimization. By the proposed method, the relationships between objects and cameras are expressed in polar coordinates easily. It is suitable for multi-camera calibration. Cameras can be calibrated with fewer points. The calibration images can be positioned according to the location of cameras. The experiment results demonstrate that the proposed method is an efficient calibration method. By the method, cameras are calibrated conveniently with high accuracy.