Isolated Chinese Sign Language Recognition Using Gray-Level Co-occurrence Matrix and Parameter-Optimized Medium Gaussian Support Vector Machine

Abstract

In order to recognize Chinese sign language more accurately, we proposed an efficient method using gray-level co-occurrence matrix (GLCM) and parameter-optimized medium Gaussian support vector machine (MGSVM). First, sign language images were acquired by digital camera or picked from video as keyframes, and then the hand shapes were segmented from background. Second, each image was resized to N × N size and converted into gray-level image. The number of intensity values in grayscale image was reduced from 256 to 8, and gray-level co-occurrence matrix was created. Third, the extracted and reduced features were sent to MGSVM; meanwhile, the classification was performed on a tenfold cross-validation. The experimental results of the 450 isolated Chinese sign language images from the 30 categories demonstrated that the GLCM–MGSVM achieved a classification accuracy of 85.3%, which was much higher than GLCM-DT (decision tree). Therefore, the GLCM-MGSVM was seen to be effective in classifying Chinese sign language.