Summary: | 碩士 === 國立成功大學 === 醫學資訊研究所 === 98 === A sign language can be completely defined by a finite set of specific gestures. This characteristic has made sign recognition an appealing application of gesture recognition. In Sign Language, a hand gesture occlusion with complex background may result a fault hand shape. To recognize a sign, a major difficulty arises because of the similar chromatic feature between the hand and background. In the presence of occlusion, it is difficult to extract the precise contour of the hand shape.
In this work, we use a MRF-based kernel feature extraction and improved level set model to track contours of the hand shape in feature space. The kernel-based feature extraction incorporates contextual and class information into the RBF kernel function to improve the performance of feature extraction. Along with the extracted features, we proposed an improved level set method with fitting and prior terms to solve the occlusion false effect.
In our experiment, the results show the applicability of the proposed method for recognizing the sign words in daily life. Table of Contents
|