Multi-view angle three-dimensional human face scanning data automatic registration method

Abstract

The utility model discloses an automatic registration method for multi-view three-dimensional face scanning data, which is characterized in comprising the following steps: coordinate axis transformation is made to the multi-view three-dimensional face scanning data through the PCA method; the shape index value of a transformed model is calculated, and a plurality of feature regions are screened using the method of threshold segmentation; region screening is restricted by region relative distribution features so as to finally position the symbolic regions such as eyes, outer canthi and nasal tip points; the symbolic regions are matched by the ICP method; and the coordinate translation and rotation transformation are made to the transformation parameters which are obtained from region registration of the whole scanning data so as to complete registration. The coarse registration of the utility model adopts the method of coordinate axis transformation, and the global features are instead of the individual point features, so error caused by symbolic point detection is reduced. The precise registration only makes registration iterative computation to the extracted symbolic regions, so computation complexity is reduced, and the automatic registration of multi-pose three-dimensional scanning data is enabled.

Description

Technical field [0001] The invention relates to a method for automatically registering multi-view three-dimensional face scan data. Background technique [0002] The registration of three-dimensional face scan data is an important step in the establishment of a three-dimensional morphological model. The registration problem includes finding the translation and rotation parameters that can correctly register the overlapping parts of the multi-view object. The object can be reconstructed from a part of the surface to obtain the object The complete surface description. For a three-dimensional human face, to obtain a complete description of the human face, at least three scan data from different perspectives (front, left, and right) need to be registered together. [0003] According to the different strategies for finding the corresponding relationship between the models, the registration methods can be divided into two categories: overall registration and piece-by-piece registration...

AI Technical Summary

Problems solved by technology

[0005] In order to overcome the shortcomings of the prior art that the accuracy rate of feature point detection is not high and cannot be used for automatic registration of multi-view 3D scan data, the present invention provides a method for automatic registration of multi-view 3D face scan data, wherein the coarse registration The method of transforming the coordinate axes is used to replace individual point features with overall features, which reduces the error caused by feature point detection; fine registration only performs registration iterative operations on the extracted feature areas, which can reduce computational complexity

Images