Gender recognition using dynamic gait energy image

This paper presents an improved gait energy image (GEI) named D-GEI. Nowadays, numerical literatures have focused on gait analysis, yet most approaches do not fully exploit the dynamic walking information, which leads to inferior performance under the appearance change and viewpoint variation. Furthermore, an improved gait energy image, D-GEI, is proposed. Firstly, we calculate and divide the GEI into dynamic region. Then the region of GEI has the logic ‘and’ operation with the sequence to get the dynamic region of frame. Finally, we calculate the weighted average of these dynamic regions to get the D-GEI. In tests, HOG characteristics of D-GEI is regarded as features representation and the tests based on the CASIA dataset are conducted, in which we select the SVM as classifier. The experimental results show that with HOG based on dynamic gait energy image, the proposed method outperforms HOG based on GEI method.