Proceeding of The 1st International Conference on Computer Science and Engineering 2014|65 Hand Contour Recognition In Language Signs Codes Using Shape Based Hand Gestures Methods Ade Silvia 1 , Nyayu Latifah Husni 2 1,2 Electrical Engginering Polytechnics of Sriwijaya 1 ade_silvia_armin@yahoo.co.id 2 latifah3576@yahoo.com Abstract— The deaf and speech impaired are loosing of hearing ability followed by disability of developing talking skill in everyday communication. Disability of making normal communication makes the deaf and speech impaired being difficult to be accepted by major normal community. Communication used is gesture language, by using hand gesture communication. The weakness of this communication is that misunderstanding and limitation, it’s due to hand gesture is only understood by minor group. To make effective communication in real time, it’s needed two ways communication that can change the code of hand gesture pattern to the texts and sounds that can be understood by other people. In this research, it’s focused on hand gesture recognition using shaped based hand algorithm where this method classifies image based on hand contour using hausdorff and Euclidian distance to determine the similarity between two hands based on the shortest range. The result of this research is recognizing 26 letters gesture, the accuracy of this Gesture is 85%, from different human hands, taken from different session with different lighting condition and different range of camera from image. It also can recognize 70% different hand contour. The different of this research from other researches is the more the objects are, the less the classification of hands size is. Using this method, hands size can be minimized. Keywords— Hand contour, Gestures, Shape-based hand gesture methods, Hausdorff distance, Euclidean distance. I. INTRODUCTION To communicate with deaf people, it is mostly used sign language or often called Indonesian Gestures Language Systems, i.e. the sign language that uses hand and finger movements. Along with advances in technology, it has developed some methods of learning (self-learning) for speech impaired and deaf people who want to learn to speak. One of them is a method developed in English by the ABC organization, whereas in Indonesian this method has not been developed. Therefore, the researchers design a learning system for the speech impaired and deaf patients through a software with expectations the patients can perform the learning through computer media [1]. Indonesian with the hand gestures pattern was developed [2] by using an artificial network, with only 69% accuracy rate values. For the encoding, it must use a PC (Personal Computer) in order to overcome the problem on resolution, besides that gestures recognition was limited only 15 words, as well as pattern recognition (hand gesture recognition) used is still static, whereas for hand gesture recognition, it is needed dynamic hand gesture recognition due to the shift patterns of the hand gesture (movement sequences). Several studies have been developed previously, as done by Rakhman et al. (2010) using the method of tracking haar classifier and classifying image data set to train with K Nearest neighbors algorithms, the system is only able to recognize 19 letters of 26 gestures, the letters that are not be able to be recognized are; M, N, S, T, A and Z. This is due to the level of similarity between the letters signs is high and it is also due to the limitation of using only the same hand image. The purposes are to design a system of sign languages encoded by hand movements in real time for speech impaired and deaf people, by using the technique of hand gesture recognition. The systems will be useful for speech impaired and deaf people as their two-ways communication. This study was developed using method 1: Feature consists of a hand contour data. Classifier based on the modified Hausdorff distance. Method 2: Feature consists of independent components of the hand silhouette. Classifier is the Euclidean distance. By using this method, the advantage of it, if compared with other studies, is that the more the number of objects (hand contours) are used, the less hand size classification is. Thus, using this method, although there are a lot of subjects to be used, the size of the hands can be minimized. II HAND SEGMENTATION Image segmentation is the process of grouping the image into several regions based on certain criteria. In this study the hand segmentation aims to extract the hand region from the background. Segmentation divides two objects, which consist of the hand and the background, but in reality, the accuracy of the segmentation will be reduced because of the presence of rings, cuffs overlap, or rope/chain watches or folds around the boundary due to the slow or strong pressure. Moreover, the depiction of the hand contour must be accurate. It is caused of the difference between the hands of different individuals. Erdem Yoruk and friends had compared two different segmentation methods, namely clustering segmentation method, which is followed by morphological operations, and segmentation based on watershed transformation. Normalization hand image involves registration of a hand image (registering), i.e. the global rotation and translation, as well as the re-orientation of the fingers along the direction of each individual standard, without any distortion of the shapes [4]. In fact, this is the most critical operation for biometry applications based on hand shapes when global features are used. There are also schemes that use only local features for example contour separates the fingers. The need for re- orientation is shown in the figure below [5][6][7].