Infrared and Laser Engineering, Volume. 50, Issue 8, 20200418(2021)
Infrared small dim target detection using local contrast measure weighted by reversed local diversity
Figures & Tables(14)
Fig. 1. (a) A sample of real IR image; (b) 3D distributions of different types of components. Here, TT represents true small target, NB represents normal background, HB represents high brightness background, EB represents complex background edge, and PNHB represents Pixel-sized Noises with High Brightness
Fig. 2. A local small image patch used for MRDLCM calculation. It is divided into 9 cells, and the cell size
Fig. 3. Different cases when the central pixel of
Fig. 5. Samples for the four IR sequences. (a) A sample for Seq. 1; (b) A sample for Seq. 2; (c) A sample for Seq. 3; (d) A sample for Seq. 4
Fig. 6. SCR results before and after MRDLCM calculation using different
Fig. 7. SCR results before and after MRDLCM calculation using different
Fig. 8. Calculation results for different types of pixels using the proposed MRDLCM_RLD algorithm. (a) Different cases when the central pixel of
Fig. 9. From top to bottom: the detection results using the proposed MRDLCM_RLD algorithm for Seq. 1, Seq. 2, Seq. 3 and Seq. 4. (a) The raw IR image samples of the four sequences; (b) The DFLCM result; (c)The RFLCM result; (d)The MRDLCM result; (e) The RLD result; (f) The MRDLCM_RLD result; (g) The threshold operation results, each connected area is regarded as a target
Fig. 10. Detection result using only MRDLCM alone for Seq. 3. (a) The raw IR image sample of Seq. 3; (b) The MRDLCM result; (c) The threshold operation result on MRDLCM, more false alarms emerge
Fig. 11. Comparisons of detection results between different algorithms, from top to down: the detection results of Seq. 1, Seq. 2, Seq. 3 and Seq. 4 using (a) DoG; (b) ILCM; (c) NLCM; (d) WLDM; (e) MPCM; and (f) RLCM
Fig. 12. ROC curves of different algorithms for (a) Seq. 1, (b) Seq. 2, (c) Seq. 3 and (d) Seq. 4
Table 1. Features of different sequences
View tableView in ArticleTable 1. Features of different sequences
Frames Image resolution Target ID Target size Target details Background details Seq. 1 300 320×240 Only 1 7×5 • Plane target.• A long imaging distance. • Located in homogeneous sky.• Keeping little motion. • Sky-Cloud background.• Heavy clutter. • Almost unchanged. Seq. 2 100 256×256 Only 1 5×5 • Truck target.• A long imaging distance. • Located in homogeneous ground. • Keeping little motion. • Ground-Tree background. • Heavy clutter.• Change slowly. Seq. 3 100 320×256 Only 1 3×3 • Plane target.• A long imaging distance. • Located in homogeneous sky. • Very small and very weak.• Keeping little motion. • Ground-Sky background. • Heavy clutter.• Almost unchanged. Seq. 4 100 256×256 Target 1 7×5 • Boat target.• A long imaging distance. • Located in homogeneous sea. • Multi targets, including moving and stationary. • Sea-Sky background. • Heavy clutter.• Almost unchanged. Target 2 5×5 Target 3 5×5 Target 4 6×6
Table 2. Characteristics of the first frame of the four sequences
View tableView in ArticleTable 2. Characteristics of the first frame of the four sequences
Target ID Cwh Cnb SCR Seq. 1 Only 1 0.3175 6.3875 0.8896 Seq. 2 Only 1 0.9043 1.1642 2.8461 Seq. 3 Only 1 0.3540 1.1734 0.1927 Seq. 4 Target 1 0.8325 1.6194 1.5707 Target 2 0.8030 1.4725 1.2457 Target 3 0.7340 1.3254 0.8545 Target 4 0.7734 1.2787 0.7848
Tools
Get Citation
Copy Citation Text
Yuanyuan Chen, Jinhui Han, Honghui Zhang, Xiaodan Sang. Infrared small dim target detection using local contrast measure weighted by reversed local diversity[J]. Infrared and Laser Engineering, 2021, 50(8): 20200418
Paper Information
Category: Infrared technology and application
Received: Nov. 2, 2020
Accepted: --
Published Online: Nov. 2, 2021
The Author Email:
© Copyright 2018-2021 | Chinese Laser Press.
All Rights Reserved 沪ICP备15018463号-20