Reconstruction of obscured objects in segmentation maps

Abstract

Reconstructing a partially obscured elliptical antenna dish image is provided. An image of an antenna dish contour with an obscured section is received, and a centroid for use as an origin reference is determined. Indices are sequentially assigned to samples along the contour starting with a sample having zero polar angle according to the origin reference. Sets of randomly selected N samples along the contour are prepared and used to perform least squares fit (LSF) to determine a cost function for each set. Cost functions with LSF residuals below a specified threshold show good quality-of-fit. An ensemble mean of the cost functions with good quality-of-fit is calculated. Outlier samples are identified according to local extrema for the ensemble mean. Samples having poor quality-of-fit of the ensemble mean between the extrema are considered outliers. Inlier samples are used to reconstruct the obscured section of the contour.