A Sinhala Finger Spelling Interpretation System Using Nearest Neighbor Classification

Dulan Manujinda Wathugala, N.D. Kodikara


Deaf people in Sri Lanka use sign language for communication and Sinhala Finger Spelling is a part of it. The aim of this study is to build a system that can interpret Sinhala finger spelling into Sinhala text. First, the positions of the fingertips and a reference point in the lower palm area are extracted from the gesture images as feature points, using a color-coded glove and a white background.

A feature vector is then constructed having several key components; visibility of fingertips, displacement of fingertips from the reference point, and angles between the fingertips. Based on the visible fingertips, each gesture image is given an index number between 0 – 31.

A prototype system to evaluate this feature vector has been developed. Training the system is done in a supervised manner. The pattern classification is done using a variation of nearest neighbor method. The system achieves recognition rates of 98% on the training set and 62% on a new set of images.