Recognition of Indonesian Sign Language Alphabets Using Fourier Descriptor Method

Feature extraction is a process to search for special features of an object that will distinguish the characteristics between one object and another. In this study, the Fourier Descriptor method was used for extracting the feature of the Indonesian Sign Language (BISINDO) images. This research has been performed by implementing four main steps, i.e., pre-processing by converting RGB images to grayscale and binary ones, and closing operation for smoothing the images; contour detection using Moore's algorithm; feature extraction implementing Fourier descriptor with 5 kinds of coefficient i.e., 2, 5, 10, 25, and 50 to represent the feature of the images; and lastly feature recognition process by computing the image similarity using Euclidean Distance. 1820 images divided into 4 kinds i.e., standard, scaled, rotated, and translated images have been tested with 130 images of training data. Based on the test result of 130 standard images for each coefficient, the best accuracy is obtained at coefficient 25 and 50 with a similar accuracy of 96.92%. In addition, the recognition performance of Fourier descriptor and Euclidean distance reached up to above 72% in average for standard and scaled images.