Quick three-dimensional human ear identification method
A fast technology for human ear recognition, applied in the field of image processing, which can solve the problems that even the point closest to the point may not be the real corresponding point, the registration point is not very reliable, and the wrong registration point is equal, etc., to achieve The effect is good, the work is stable, and the operation speed is improved.
- Summary
- Abstract
- Description
- Claims
- Application Information
AI Technical Summary
Problems solved by technology
Method used
Image
Examples
Embodiment 1
[0045] The data used is from the University of Notre Dame's 3D human ear database (the University of Notre Dame public data set, UND), which is the largest publicly available 3D human ear database so far. It has a total of 830 datasets from 415 individuals. All data sets are obtained by scanning the left face with a Minolta 910 laser scanner, including depth images of X, Y, and Z coordinates and corresponding RGB color images, with a resolution of 640×480. The hardware environment of the system is Intel(R) Xeon(R) quad-core processor, 2.33GHz, memory 4G. The software environment is Matlab R2008a. Principle of human ear recognition based on PCA:
[0046] PCA is one of the most widely used feature extraction methods. It is a statistical method and has been widely used in signal processing, pattern recognition, digital image processing and other fields. The basic idea of PCA is to extract the main features (principal elements) of the original data distribution in space, redu...
Embodiment 2
[0061] The process of human ear extraction is as follows:
[0062] 1. Ear hole detection, using the ear hole detection method to find the position of the human ear in the depth image of the side face;
[0063] 2. Triangulation, take out the points within the range of 102×142 around the ear hole, construct a triangular mesh, use the following mesh generation method: add a hypotenuse to any four adjacent points on the X-Y coordinate plane 2 triangles, but any adjacent two hypotenuses are not parallel, some vertex data on the grid is missing, called invalid points, remove the triangle whose vertices are invalid points, and then calculate the actual side length of the triangle in three-dimensional space , mark the triangle whose maximum side length is greater than 5mm as an untrusted triangle;
[0064] 3. Coordinate centralization, by calculating the mean value of point coordinates in a small area around the ear hole, the three-dimensional coordinates of the ear hole are obtained...
Embodiment 3
[0076] Using the automatically extracted human ears, respectively, using the depth image R Z , the maximum principal curvature image k 1 , the minimum principal curvature image k 2 and their combination to train the PCA projection base on 415 prototype ears, retaining 80% of the energy, and then use the Euclidean distance measure, cosine measure and Tanimoto measure to perform nearest neighbor classification. The experimental results are shown in Table 2.
[0077] Table 2 The relationship between the recognition rate of the PCA method and the images used and the similarity measure
[0078]
[0079] It can be seen that there is no significant difference between the cosine measure or the Tanimoto measure, but both are higher than the Euclidean distance measure. When using one image alone, the minimum principal curvature outperforms the maximum principal curvature, which in turn outperforms the depth image. In the combined image, the combination of depth image and minimum p...
PUM
Login to View More Abstract
Description
Claims
Application Information
Login to View More