Abstract
Fast and stable detection of dim and small infrared (IR) targets in complex backgrounds has important practical significance for IR search and tracking system. The existing small IR target detection methods usually fail or cause a high probability of false alarm in the highly heterogeneous and complex backgrounds. Continuous motion of a target relative to the background is important information regarding detection. In this article, a low-rank and sparse decomposition method based on greedy bilateral factorization is proposed for IR dim and small target detection. First, by analyzing the complex structure information of IR image sequences, the target is regarded as an independent sparse motion structure and an efficient optimization algorithm is designed. Second, the greedy bilateral factorization strategy is adopted to approximate the low-rank part of the algorithm, which significantly accelerates the efficiency of the algorithm. Extensive experiments demonstrate that the proposed method has better detection performance than the existing methods. The proposed method can still detect targets quickly and stably especially in complex scenes with weak signal-to-noise ratio.
Highlights
I NFRARED search and tracking system is a passive detection system and is widely used in video target monitoring and other fields
local contrast measure (LCM) proposes a local contrast descriptor to measure the difference between the current position and its neighborhood to enhance the target while suppressing the background, so as to improve the image signal-to-noise ratio (SNR) [4]
The time information of image sequences is combined with the theory of low-rank matrix decomposition, and the IR small target is regarded as a special sparse noise component of complex background noise, which is modeled by Gaussian mixture and Markov random domain [37]
Summary
I NFRARED search and tracking system is a passive detection system and is widely used in video target monitoring and other fields. Typical single-frame detection methods include tophat filtering [1], max-mean/max-median filtering [2], 2-D least mean square adaptive filtering [3], etc Most of these methods are based on morphological filtering to detect targets by suppressing the background. LCM proposes a local contrast descriptor to measure the difference between the current position and its neighborhood to enhance the target while suppressing the background, so as to improve the image SNR [4]. Huang et al proposed an IR small target detection method based on density peak search and maximum gray area growth, where the target and background were segmented by selecting seed growth points [12] These methods achieve better detection performance in complex scenes.
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Free) 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
