Small infrared target detection via multiscale kurtosis maps fusion

Wang, He; Xin, Yunhong

doi:10.1049/el.2020.0701

Cited by 2 publications

(1 citation statement)

References 14 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…However, when faced with a real scene where the target size is unknown, it easily leads to target missed or false positives. Wang et al [ 35 ] proposed a multiscale kurtosis map fusion method to detect small targets, which can effectively match the target size and shape by selecting multiscale windows and achieves high accuracy and adaptable detection for infrared small targets. However, when encountering pixel-sized noises with high brightness, false detection will still occur.…”

Section: Introductionmentioning

confidence: 99%

Infrared Small Target Detection Based on Multiscale Kurtosis Map Fusion and Optical Flow Method

Xin

Cao

Xiao

et al. 2023

Sensors

Self Cite

View full text Add to dashboard Cite

The uncertainty of target sizes and the complexity of backgrounds are the main reasons for the poor detection performance of small infrared targets. Focusing on this issue, this paper presents a robust and accurate algorithm that combines multiscale kurtosis map fusion and the optical flow method for the detection of small infrared targets in complex natural scenes. The paper has made three main contributions: First, it proposes a structure for infrared small target detection technology based on multiscale kurtosis maps and optical flow fields, which can well represent the shape, size and motion information of the target and is advantageous to the enhancement of the target and the suppression of the background. Second, a strategy of multi-scale kurtosis map fusion is presented to match the shape and the size of the small target, which can effectively enhance small targets with different sizes as well as suppress the highlighted noise points and the residual background edges. During the fusion process, a novel weighting mechanism is proposed to fuse different scale kurtosis maps, by means of which the scale that matches the true target is effectively enhanced. Third, an improved optical flow method is utilized to further suppress the nontarget residual clutter that cannot be completely removed by multiscale kurtosis map fusion. By means of the scale confidence parameter obtained during the multiscale kurtosis map fusion step, the optical flow method can select the optimal neighborhood that matches best to the target size and shape, which can effectively improve the integrity of the detection target and the ability to suppress residual clutter. As a result, the proposed method achieves a superior performance. Experimental results on eleven typical complex infrared natural scenes show that, compared with seven state-of-the-art methods, the presented method outperforms in terms of subjective visual effect, as well as some main objective evaluation indicators such as BSF, SCRG and ROC, etc.

show abstract

Section: Introductionmentioning

confidence: 99%