Research On Keypoint Detection Algorithm Of 3D Mesh Model

Due to the rapid development of computer hardware devices,3D models have been widely used in life.The 3D mesh is the main expression of the 3D model due to its clear and intuitive form.The keypoint of the 3D mesh is the simplest feature element in geometry.It not only saves a lot of time and memory for the storage of the mesh,but also shows the main features of the whole mesh with a small space,and it is for object recognition.Practical applications such as matching and retrieval play the most fundamental role.Therefore,3D mesh keypoint detection has always been a research hotspot.3D keypoint detection has been studied since more than a decade ago,and many algorithms have appeared so far.But the most effective method is the combination of local features and global features.This method not only captures local saliency but also captures global saliency.In order to more accurately detect keypoints on the 3D model,this thesis uses a combination of local features and global features to study.This thesis mainly proposes a new method to detect the saliency of the mesh,which attempts to be consistent with human perception.Unlike methods that operate in the spatial domain,the method captures information corresponding to the significance of the mesh in the transform domain.In the past,methods relying on central surround local operators tend to capture local saliency,and the transform mesh saliency method proposed in this thesis outputs a saliency map,which captures the main keypoints of global significance.In this thesis,the saliency of the mesh is combined with the transformation analysis method,and the corresponding calculation model is applied to the 3D mesh keypoint detection problem.This is a perception-based measurement method used to measure the importance of local regions on a 3D mesh.The algorithm in this thesis combines global considerations by taking advantage of the transform properties of the mesh,unlike most existing methods,which are usually based on local geometric cues.This thesis first considers the properties of the logarithmic Laplacian transform of the mesh,using those frequencies that show differences from the expected behavior to capture saliency in the transform domain.Information about these frequencies is then considered in a spatial domain of multiple spatial scales to locate salient features and give the final salient regions.In this thesis,a general benchmark is used to evaluate the proposed algorithm.Firstly,the visual demonstration is compared with the real keypoints.It is concluded that the keypoint of the algorithm detection is closer to the real keypoint.In addition,three statistical metrics are used for quantitative analysis,namely false negative error,false positive error and weighted miss error.It can be seen from the three error graphs that the error of the keypoint detection algorithm based on transform saliency is small.It is proved that the 3D keypoint detection algorithm based on transform saliency can detect the correct keypoints.