Structuring visual information for person detection in video: Application to VIDTIMIT database
Beldi Makrem, Zied Lachiri · 2016
The current paper suggests an initial segmentation system, detection and grouping of visual information based on faces to facilitate the use, the handling and validation of the VIDTIMIT database. In order to implement this basis, two methods of pretreatment for each face are used. Then faces are detected with the Viola and Jones algorithm based on weak type descriptors Haar and grouped by new faces and not by Plans initially provided in this database. The first is based on the variation or change of the luminance. The second is based on the normalization histogram rank (histogram remapping). In order to validate this methods, to use a classification system to compare the influence of both techniques on classification rate based on the PCA and LDA principles.