Abstract
Recently, pattern recognition methods and data discrimination by them become more and more important in the engineering fields. For example, digital cameras can detect human faces by recognizing the shape of nose or mouth in the sight. Thus, the pattern recognition is applied to various areas such as image processing and speech recognition. Linear Discriminant Analysis (LDA) and Support Vector Machine (SVM) are usually used for the pattern recognition methods. However, these methods do not consider the geometric feature in the distribution of the original data. We introduce the method that approximates original data distribution without losing the geometric features such as line or circle by using Conformal Geometric Algebra (CGA). This paper proposes a pattern recognition method reflecting the geometric feature of the data based on the approximation method. This paper applies the proposed method to artificial and benchmark data and studies the effects of the proposed method in terms of the discriminant accuracy comparing with Quadratic Discriminant Analysis (QDA).
