A Comparative Analysis of Segmentation Algorithms for Hand Gesture Recognition Rohit Kumar Gupta Researcher Innovation Labs, TCS Kolkata, India rohit3.gupta@tcs.com Abstract—One of the major challenges in hand gesture recognition is to segment the hand region effectively in varying background and changing lighting condition. The aim of this work is to evaluate different segmentation processes specific to hand gesture recognition. Motivation of this research is the lack of direct and detailed comparison of those algorithms in different working conditions. In this paper six highly efficient algorithms used for hand region segmentation are compared and the results are presented using different types of input images ranging from complex background to simple background, closer views to distant views and textured images to simple images. Keywords-component; Segmentation, Histogram, Clustering, Gaussian Modeling I. INTRODUCTION In large number of image processing and computer vision applications, segmentation plays a fundamental role as the first step of low-level processing. Segmentation decomposes an image into meaningful parts by retaining the region of interest (ROI) and filtering out the non-relevant regions. Some of the widely used approaches are either finding edges or boundaries or by determining disjoint and homogeneous regions. Good segmentation should exhibit certain desirable characteristics as stated by Haralick and Shapiro [1]. They mentioned that regions should be uniform and homogenous with respect to gray tone and texture. Nevertheless, according to Fu et al. [2], the image segmentation problem is basically one of psychophysical perception, and therefore not susceptible to a purely analytical solution. Another commonly used approach for segmentation is Background Subtraction (BGS) which is commonly used to segment foreground objects using video streams. The popularity of BGS largely comes from its computational efficiency, which allows applications such as video surveillance, traffic monitoring and virtual reality games. In this study, segmentation using background subtraction and other color based techniques has been considered which is becoming increasingly important in many applications. For instance, hand gestures are now used for in-flight entertainment, medicinal uses [12], car-infotainment systems [11] and for many human computer interactive games. This paper has been divided into 4 Sections with Section 2 discussing different Background subtraction methods in detail and Section 3 discussing Color based approaches for segmentation of static hand gestures. In each Subsection observations of each algorithm in three different case scenarios (Fig. 1) are given. Figure 1. Sample Input Images with simple background, bad illumination and complex background II. BACKGROUND SUBTRACTION TECHNIQUES The major task of BGS is to build an explicit model of the background [3][4][6]. Segmentation is then performed to extract foreground objects by calculating the difference between the current frame and the background model. A good BGS algorithm should be robust to changing illumination conditions, able to ignore the movement of small background elements, and capable of incorporating new objects into the background model. BGS algorithms typically follow the data flow diagram illustrated in Fig. 2. The survey by Radke et al. [17] provides a useful summary of preprocessing techniques. Figure 2. Background subtraction steps Piccardi [7] investigated the computational complexity, memory requirements, and theoretical accuracy of seven BGS algorithms. A comprehensive comparison of ten different BGS algorithms was conducted by Toyama et al. [13]. BGS can be further classified to Recursive and Non- recursive techniques. Recursive techniques include Running Gaussian average (RGA), Gaussian Mixture Model (GMM), Adaptive GMM (AGMM), Approximated median filtering (AMF) [16]. Non-recursive techniques use a buffer of some video frames and estimate a background model based solely on the statistical properties of these frames. It includes Median filtering and Eigenbackgrounds [19], HOG and 2011 Third International Conference on Computational Intelligence, Communication Systems and Networks 978-0-7695-4482-3/11 $26.00 © 2011 IEEE DOI 10.1109/CICSyN.2011.57 231