On the detection of tracks in spectrogram images

• Published in: Pattern recognition Vol. 46; no. 5; pp. 1396 - 1408
• Main Authors: Lampert, Thomas A., O'Keefe, Simon E.M.
• Format: Journal Article
• Language: English
• Published: Kidlington Elsevier Ltd 01.05.2013; Elsevier
• Subjects: Active contour; Applied sciences; Computer Science; Computer Vision and Pattern Recognition; Detection, estimation, filtering, equalization, prediction; Energy minimisation; Exact sciences and technology; Feature extraction; Image detection; Image processing; Information, signal and communications theory; Noise levels; Pattern recognition; Potential energy; Remote sensing; Shape; Signal and communications theory; Signal processing; Signal representation. Spectral analysis; Signal to noise ratio; Signal, noise; Spectrogram; Spectrograms; Statistical pattern recognition; Structural pattern recognition; Telecommunications and information theory; Statistical pattern recognition; Energy minimisation; Structural pattern recognition; Active contour; Remote sensing; Spectrogram; Potential energy; Performance evaluation; Target tracking; False positive; Image processing; Refinement method; Frequency domain method; Pattern recognition; Shape detection; Algorithm; Spectrum analysis; Signal processing; Feature extraction; Signal analysis; Edge detection; Non contact measurement; Signal to noise ratio; Remote Sensing; Energy Minimisation; Structural Pattern Recognition; Statistical Pattern Recognition; Active Contour
• ISSN: 0031-3203; 1873-5142
• DOI: 10.1016/j.patcog.2012.11.009

This paper proposes an active contour algorithm for spectrogram track detection. It extends upon previously published work in a number of areas, previously published internal and potential energy models are refined and theoretical motivations for these changes are offered. These refinements offer a marked improvement in detection performance, including a notable reduction in the probability of false positive detections. The result is feature extraction at signal-to-noise ratios as low as −1dB in the frequency domain. These theoretical and experimental findings are related to existing solutions to the problem, offering a new insight into their limitations. We show, through complexity analysis, that this is achievable in real-time. ► Commonly employed continuity and curvature measures are not suitable, and increase false positive detection rates. ► A curvature measure based on geometric properties of the track accurately models various track appearances. ► Track detection methods based upon individual pixels are unreliable at low SNRs. ► The proposed active contour algorithm effectively detects tracks of varying appearances. ► The line location accuracy metric is flawed and can rate good and bad detections similarly.