Projection-variance algorithms for constructing directional fields

A method for constructing fast projection-variance algorithms based on convolution algorithms for calculating variances of Radon projections is proposed. It is shown that the use of bilinear interpolation of the convolution kernel and parabolic interpolation of quantized estimates of directional fields increases the accuracy of the algorithms by more than a factor of 2. Two methods for constructing directional fields on the basis of Radon projections are considered and compared. Examples are given showing the computational efficiency, accuracy, and noise resistance of the proposed algorithms.