n method of Angular Voting Approach in n-sphere space of multi-dimensional features considering sensitivity

摘要

The present invention relates to an n-th order equiangular boring method of a multidimensional feature taking sensitivity into consideration, ) Into n-spehere-based spherical space coordinates ( , Transforming the input size (r) of the spherical space coordinates into a subspace Calculating a respective sub-input magnitude (r _k); and the partial differential of the change on the input of the phase (θ _k) of each sub-area of the spherical space coordinates sensitivity (S _k) for distribution, the sensitivity (S _k , and distributing the bowing value for each phase bin bin by reflecting the sub-input phase r _k distributed in the subspace and the phase θ _k of the subspace. According to the multi-dimensional feature-based neural network based on the sensitivity, the robust feature vector coding that can obtain the recognition performance for a small number of databases is provided, and the recognition and detection performance is improved when applied to HAD or HAR .