Hand gesture is a nonverbal communication which is very useful to the deaf and mute. It is also used as an alternative way to communicate with computers. Hand gesture recognition has a wide range of applications such as recognizing of sign language, interfaces for human-computer interaction, robot control, machine vision, smart surveillance, computer games, keyboards and mice replacement. This paper described a hand gesture recognition system, a vision-based approach, to recognize static hand gesture images using Kohonen Self-Organizing Map (SOM), an artificial neural network which learns to classify data without supervision. A set of 29 hand gesture images representing letters of the alphabet, enter, space and backspace keys were captured using a CMU camera. The images were cropped using a photo editor and the edited images were converted to grayscale using the MATLAB software. These images in 1D form were used as training set for the Kohonen Self-Organizing Map. After the unsupervised training, the system was tested using 29 actual hand gestures and 10 trials for each gesture. The system achieved an average of 91% accuracy with only 9% error. The system’s recognition accuracy may be further improved by increasing the number of epochs in the training phase, experimenting to find a better learning rate, using a high-resolution camera to capture the image more precisely to minimize the amount of background noise resulting to a more defined input feature vector to be fed to the SOM.

Keywords : Artificial Neural Network, Image Processing, Kohonen Self-Organizing Map, Vision-Based Approach.