ASL Fingerspelling Classification

Evaluating the Effectiveness of Various Methods of Image Transformation in American Sign Language (ASL) Fingerspelling Image Classification

There are just under one million Americans who are functionally deaf, many of whom use a form of sign language to communicate. The most common dialect of sign language in the United States is American Sign Language (ASL), which includes a fingerspelling alphabet to convey meaning through spelling out words. Although many common words have shorter signs, fingerspelling is still very useful for words that don’t have other signs, like proper nouns, or for words that the speaker does not know another sign for. Previous studies have worked to classify ASL fingerspelling images, however, we look to evaluate the effectiveness of using different tools, including background subtraction and hand identification. Each transformation method was applied to the images and a classifier will be trained on each dataset. The effectiveness of each transformation method in aiding classification will then be evaluated. In the future, we hope to apply this technology to real-time sign language translation, including signs outside of the sign language alphabet, evaluating the different image transformation methods on a larger scale. We hope that this research will help to bridge part of the communication gap faced between deaf and hearing people.