A novel gait contours segmentation algorithm

Effective extraction of human area from human-body moving gait image sequences is of great concern to human-body moving gait recognition. A new segmentation algorithm was proposed, the concept of the time-domain was afresh defined, the inverse transformation of the wavelet was unnecessary. The two value result which was gained by combining background subtraction and symmetric frame difference performed moving estimation, it was the new method of time-domain segmentation; the first order wavelet transform was used to the current frame image, used high order linear interpolation to extend the LL weight of wavelet transformation as the same size as the current frame image, watershed segmentation algorithm was used to divide the LL weight image which was extended into many closed and non-overlapping regions, it was space-domain segmentation; used spatio-temporal combination algorithm, marked the result of space-domain segmentation casting shadow to the result of time-domain segmentation, then gained the accurate area of human-body. The experiments were made on the NLPR gait database and the results proved that the algorithm can abstract the whole area of human-body precisely, owned the good robust noise and flexibility, and eventually improved the performance of gait recognition.