Contours extraction in grayscale images of sign language interpreter

• Authors: Roumen Kountchev;Vladimir Todorov;Roumiana Kountcheva
• Venue: MMACTEE'08 Proceedings of the 10th WSEAS International Conference on Mathematical Methods and Computational Techniques in Electrical Engineering
• Year: 2008

This work presents one new approach for adaptive extraction of the contours of the sign language interpreter in video sequences. The aim of the processing is to obtain clear contour images, which retain the understandability of the interpreted signs and together with this, ensure very efficient compression of the processed video sequences. In the paper is presented the algorithm used in the consecutive stages of the image processing illustrated with some experimental results. The algorithm is based on two-stage filtration (noise filtration with local adaptive digital filtration, followed by uneven illumination correction with 2D linear digital filtration). The processing then continues with image segmentation. The presented approach is of great importance for video sign-language presentations, because in result of the usually used inter-frame compression their understandability is not so good, if they comprise quick movements of interpreter's hands. In order to avoid this, the new approach is based on the assumption that the color video sequences are substituted with similar ones, which contain the extracted contours only. The understandability of the represented signs is retained, because in this case instead of inter-frame lossy compression, is used intra-frame lossless compression. The obtained compression ratio is very high, which permits real-time implementations of the presented approach.