In this paper, we propose a new approach to dynamic signature verification as a biometric feature. Such a signature is represented by its shape and the trajectories describing the dynamics of signing, which are difficult to forge. The proposed approach utilizes signature regions. These are sets of discretized points of the shape and dynamic trajectories of the signature, as well as various combinations thereof. The points originate from three parts of the signature: the initial, middle, and final segments, with the same number of points in each region. This facilitates the unification of the neural network structure for their processing, eliminates the problem of analyzing varying numbers of discretization points from signatures of different users, and refers to the concept of our previous research on dynamic signature partitioning. In the simulations, we used an Long Short-Term Memory network for signature verification, and the achieved accuracy was very high.