Author Affiliations
1North China Research Institute of Electro-optics, Beijing 100015, China2Beijing Vacuum Electronics Research Institute, Beijing 100015, Chinashow less
Fig. 1. Blind element 15×15 neighborhood gray distribution on clean background
Fig. 2. Flow chart of the proposed algorithm
Fig. 3. Local extremum operator
Fig. 4. Defective pixel location based on extremum operator and three-layer pyramid
Fig. 5. FPGA logic implementation flow of the proposed algorithm
Fig. 6. Timing simulation result of hardware implementation
Fig. 7. Target and defect element sequence set in multiple scenarios
Fig. 8. Comparison of FPGA implementation using different methods. (a) Original image output by the detector; (b) Multi-scale adaptive median filter; (c) Local contrast method based on saliency; (d) Structural elements and 3σ criterion; (e) Proposed algorithm
Fig. 9. Diagram of the performance improvement for point target detection. (a) Multiscene infrared point target detection; (b) Target neighborhood; (c) 3 D image of grayscale distribution of point target neighborhood; (d) Target neighborhood with the proposed algorithm; (e) 3 D image of the grayscale distribution of point target neighborhood with the proposed algorithm
Type | Gray distribution of spatial | Gray distribution of temporal | Over hot pixel | | | Dead pixel | | | Flickering pixel | | | Defect pixel cluster | | | Point target | | | Background | | |
|
Table 1. Comparison of temporal and spatial gray distribution characteristics of defect element, target and clean background
Resolution | Device
materials
| Band | Frame-rate/
fps
| Integrated-time/
ms
| 640×512 | MCT | Mid-wave | 100 | 2.8-7.5 | 640×512 | MCT | Long-wave | 300 | 0.3-1.2 | 1 k×1 k | MCT | Mid-wave | 75 | 3-8 | 256×256 | IS | Mid-wave | 75 | 0.5-3.6 | 64×64 | IS | Long-wave | 500 | 0.2-0.5 | 640×512 | IS | Long-wave | 100 | 0.8-5 | 320×256 | SLs | Long-wave | 75 | 0.5-1.8 |
|
Table 2. Comparison of BSF of different algorithms
Algorithm | Isolated
blind
pixel
| Isolated
flickering pixel
| Defect
pixel
cluster
| Timing
summary/ms
| Existing problems | AMF | Valid | Valid | Valid | 0.325 | Background detail information are lost; Dim target will be removed | LCM | Valid | Invalid | Invalid | 0.830 | Dim target will be removed | SE-3σ | Valid | Invalid | Invalid | 3.421 | Dim targets may be removed;Algorithm running time is a little long | OCSVM | Valid | Invalid | Invalid | -- | Difficult to implement on FPGA; Dim target may be removed | DDABS | Valid | Valid | Invalid | 0.326 | Unable to adjust time-consuming flexibly | Proposed | Valid | Valid | Valid | 0.329 | N/A |
|
Table 3. Effects and problems of different algorithms for eliminating isolated pixel, flickering pixel and defective pixel clusterss
Algorithm | Clean background | | Ground background | | Complex cloud background | SCRg | BSF | PFg | DAR | SCRg | BSF | PFg | DAR | SCRg | BSF | PFg | DAR | Proposed | 1.21 | 1.36 | 2.34 | 0.98 | | 1.45 | 1.63 | 8.21 | 0.95 | | 3.15 | 1.92 | 8.92 | 0.95 | AMF | 0.12 | 1.81 | 6.43 | 0.52 | 0.31 | 2.72 | 3.24 | 0.49 | 0.32 | 3.32 | 2.93 | 0.48 | LCM | 1.16 | 1.23 | 1.89 | 0.83 | 0.84 | 1.28 | 1.42 | 0.76 | 0.78 | 1.76 | 1.39 | 0.73 | SE-3σ | 1.19 | 1.32 | 2.18 | 0.87 | 1.34 | 1.36 | 2.32 | 0.85 | 1.45 | 1.52 | 1.73 | 0.81 | OCSVM | 1.03 | 1.32 | 1.35 | 0.76 | 1.16 | 1.29 | 1.24 | 0.74 | 1.13 | 1.47 | 1.16 | 0.77 | DDABS | 1.21 | 1.36 | 1.92 | 0.91 | 1.42 | 1.54 | 4.21 | 0.88 | 2.38 | 1.74 | 5.21 | 0.82 |
|
Table 4. Comparison of detection performance of different algorithms in three kinds of scenes