Low-SNR Sea-sky-line Extraction using Fuzzy Voting by Local Line Segment

This paper proposes a novel sea-sky-line extraction method using the “hypothesize-and-verify” paradigm. In the hypothesizing step， a novel line segment Hough transform is proposed to detect some likely sea-sky-line . Firstly， the local line segment is used to estimate line parameters. The theoretical deduction proves that the estimating error variance is much less than the average of estimating error variances of all edge points， which are estimated independently. Secondly， votes are cast in the parameter space fuzzily by local line segments to obtain global line segment clustering. Finally， peaks are detected to generate likely sea-sky-line hypotheses. In the verifying step， three kinds of new features are proposed to describe the difference between sea-sky-line and sea clutter. And a SVM classifier is used to recognize the sea-sky-line . The proposed approach can extract the low-SNR sea-sky-line correctly and has high correct recognition rate. Simulation results and experimental results demonstrate the effectiveness of the proposed approach.