Clustering Analysis Of Chernoff Faces Based On The V-system

V-system is an orthogonal complete function on L₂[0,1], composed of piecewise polynomial, which includes not only continue functions, but also discontinue functions. The zero order V-system is just Haar wavelet. A complex geometric model can be expressed with finite terms of the V-series accurately. V-system has properties of multi-resolution and local support, which play an important role in computer graphics, computer vision, CAD/CAM, medical imaging, scientific calculation and so on. In this paper, the characteristic of V-system will be applied to cluster the Chernoff faces.The Chernoff face is a classical method to display multidimensional data graphically in multivariate statistics and an effective data visualization technology. The Chernoff face reflects the information features of multi-variable data, and be used as an important tool representing high-dimensional data. In 1973, Chernoff, an American statistician, proposed Chernoff faces for clustering analysis firstly, and multi-variable expression in the plane is increasingly concerned from then on. The presentation of Chernoff faces has developed the graphical multivriate statistical Analysis. And it is an important mean for people to understand the multi-dimensional space visualizingThe central work of this paper is as follows:(1)The Chernoff faces are reconstructed accurately via finite terms of the V-series. However, Fourier function systems, continuous wavelet and almost all well-known classical orthogonal continuous systems have inevitably caused Gibbs phenomenon when they are used to reconstruct the geometric model with discontinuous information.(2)The V-descriptor is used to cluster analysis of the Cheronff faces. By quantifying the overall features of the Chernoff faces, we offer a new programmed clustering method for Chernoff faces. It can avoid misjudgment caused by human eyes. Especially when the number of data sets processing is very large, this approach is even more advantages. The concrete examples show that the clustering algorithm is a simple, fast, effective, and the clustering results are in line with the clustering results of the SAS statistical software.