Shock Graph Features and Decision Tree Classifier in Human Posture Recognition Nooritawati Md Tahir* , Aini Hussain, Salina Abdul Samad and Hafizah Husain Signal Processing Research Group, Dept. of Electrical, Electronics and Systems Faculty of Engineering, Universiti Kebangsaan Malaysia 43600 Bangi, Selangor DE, MALAYSIA. Tel : 603-89216035/22 Fax : 603-89216146 . *Corresponding author: norita_tahir@yahoo.com or norita@vlsi.eng.ukm.my Abstract. In this paper, we endeavour in realizing the potential of our distinct version of shock graph (SG) as features for human shape representation. The developed technique involves the following procedures of skeletonization and thinning. Next, for modelling the human shape, end points and junction points are systematically detected using our novel pruning procedure. Six principal points, which consists of two junction points and four end points for each SG, were determined and preserved. Three feature vectors that epitomize the human shape were extracted and these features represent the parameters of length l 1 , l 2 , and l 3 . These human shape were then classified into four categories of human postures namely standing frontage (SF), standing sideways (SS), non standing frontage (NF) and non standing sideways (NS). These three feature vectors acted as inputs to the classifier. We further anticipate to evaluate the potential of adopt Classification and Regression Tree (CART) as classifier. Since the three feature vectors are continuous data, it needs to be discretized to broaden the potential of the method. Therefore, these continuous data can be easily utilized after being discretized. Initial results of the experiments are encouraging which suggested that our method can efficiently be applied for posture classification using DT. Index Term: Shock Graph, Human Posture, Decision Tree, C ART 1. Introduction The advent of computers has increased the demand for practical applications of pattern recognition, which in turn set new demands for further theoretical developments. The need for information handling and retrieval has pushed pattern recognition to the high edge of today’s engineering applications and research. Pattern recognition is an integral part in most machine intelligence systems built for decision making tasks such as computer vision, biometric, document classification and bioinformatics. The process of pattern recognition can be divided into three principal steps namely data acquisition, feature selection and classification. Here, skeleton transform, of which the Medial Axis Transform (MAT) is the most popular (1),(2),(3),(4), will be evaluated as useful features for representation and modeling of human posture. This process involves reducing foreground regions in a binary image to a skeletal remnant that largely preserves the extent and connectivity of the original region while eliminating most of the original foreground pixels. One way to think about the skeleton is as the loci of centers of bi-tangent circles that fit entirely within the foreground region being considered as illustrates in Figure 1. This is an example of a skeleton for the triangle shape. The structure of this paper is as follows. The format and methods are discussed in Section 2. Section 3 describes the test we performed and Section 4 evaluates the preliminary results achieved. Finally, in Section 5, we conclude our findings. 2. Methodology In this work, the human shapes will be represented by their unique shock graph (SG). SG is effective because it offers simple and compact representation of a shape whilst preserving most of the topological and size elements of the original shape. For instance, we can get an idea of the shape length simply by considering just the end points of the skeleton and finding the maximally separated pair of end points on the skeleton. Similarly, we can distinguish many different shapes from one another based on the number of ‘end points and junction points’ encompassed in the shapes. The definition of end points and junction points are recall here as: A point of a one-pixel width digital curve is an end point if it has a single pixel among its 3x3 neighbourhood. A point of a one-pixel width digital curve is defined as a junction point if it has more than two curve pixels among its 3x3 neighbourhood. Figure 1: Skeleton of a triangle defined in terms of bi-tangent circles (Thick dark line segments mark the skeleton.) Proceedings of the International Conference on Electrical Engineering and Informatics Institut Teknologi Bandung, Indonesia June 17-19, 2007 B-30 ISBN 978-979-16338-0-2 195