Line detection in strongly noise-corrupted images

Abstract

A novel method for detection of thin lines in strongly noise-corrupted images is presented. The method can be applied e.g. for guide wire detection in X-ray fluoroscopy sequences. In a first step, the responses of a set of angular separable filters are calculated and the maximum of the filter responses is taken. Application of a subsampling scheme allows to reduce the computational effort. In a second step, a flood filling algorithm is used to eliminate worm-like artifacts arising from random line structures in the noise. Depending on the line curvature, signal-to-noise ratio (SNR) improvements of 27 dB have been achieved.