Author Affiliations
1Key Laboratory of Intelligent Infrared Perception, Shanghai Institute of Technical Physics, Chinese Academy of Sciences, Shanghai 200083, China2University of Chinese Academy of Sciences, Beijing 100049, Chinashow less
Fig. 1. Schematic of point source method
Fig. 2. Ideal two-dimensional Gaussian distribution. (a) Three-dimensional view; (b) sectional view
Fig. 3. Influence of partition parameter P on partition
Fig. 4. Influence of partition parameter P on MSE of MTF (a=0.8). (a) MSE of image with noise standard deviation of 30 under various denoising methods; (b) relationship between parameter P and MSE under different noise standard deviations
Fig. 5. Influence of partition parameter a on data smoothing
Fig. 6. Influence of smoothing parameter a on MSE of MTF (P=0.3). (a) MSE of image with noise standard deviation of 30 under various denoising methods; (b) relationship between parameter a and MSE under different noise standard deviations
Fig. 7. Sequence of out-of-focus images. (a) image 1; (b) image 2; (c) image 3; (d) image 4; (e) image 5; (f) image 6
Fig. 8. Three-dimensional images and two-dimensional profiles of different denoising methods (noise standard deviation is 20). (a) Original point source image; (b) add noise point source image; (c) mean filtering; (d) median filtering; (e) wavelet filtering; (f) proposed method
Fig. 9. MTF measured after different denoising methods
Fig. 10. MSE of MTF under different denoising methods
Fig. 11. PSNR of images under different denoising methods
Fig. 12. SSIM of images under different denoising methods
Fig. 13. Sequences of captured out-of-focus source images. (a) -2.5 mm; (b) -2.0 mm; (c) -1.5 mm; (d) -1.0 mm; (e) 0 mm; (f) 1.0 mm; (g) +1.5 mm; (h) +2.0 mm; (i) +2.5 mm
Fig. 14. MTF measured after different denoising methods (defocused amount: -2.5 mm)
Standard deviationof noise | Performance | Noiseimage | Proposedmethod | Mean filteringmethod | Medium filteringmethod | Wavelet filteringmethod |
---|
| MSE | 0.0107 | 0.0023 | 0.0054 | 0.0060 | 0.0047 | 5 | PSNR | 39.5126 | 41.5314 | 41.0826 | 40.9626 | 41.1824 | | SSIM | 0.9502 | 0.9940 | 0.9829 | 0.9789 | 0.9850 | | MSE | 0.0205 | 0.0019 | 0.0110 | 0.0133 | 0.0087 | 10 | PSNR | 32.6774 | 35.9853 | 35.1366 | 35.2308 | 35.4256 | | SSIM | 0.7662 | 0.9814 | 0.9367 | 0.9236 | 0.9571 | | MSE | 0.0270 | 0.0040 | 0.0155 | 0.0173 | 0.0127 | 15 | PSNR | 28.2794 | 32.4906 | 31.6207 | 32.0162 | 31.9966 | | SSIM | 0.5133 | 0.9616 | 0.8688 | 0.8452 | 0.9209 | | MSE | 0.0321 | 0.0061 | 0.0193 | 0.0203 | 0.0159 | 20 | PSNR | 25.1575 | 30.0172 | 29.1058 | 29.9099 | 29.5053 | | SSIM | 0.3225 | 0.9391 | 0.7883 | 0.7601 | 0.8740 | | MSE | 0.0330 | 0.0096 | 0.0210 | 0.0223 | 0.0186 | 25 | PSNR | 22.8968 | 28.0722 | 27.1704 | 28.4650 | 27.5833 | | SSIM | 0.2119 | 0.9148 | 0.7068 | 0.6848 | 0.8236 | | MSE | 0.0338 | 0.0098 | 0.0230 | 0.0242 | 0.0200 | 30 | PSNR | 21.1746 | 26.4851 | 25.5809 | 27.3332 | 25.9853 | | SSIM | 0.1484 | 0.8875 | 0.6280 | 0.6173 | 0.7682 | | MSE | 0.0262 | 0.0056 | 0.0159 | 0.0172 | 0.0134 | Mean value | PSNR | 28.2831 | 32.4303 | 31.6162 | 32.3196 | 31.9464 | | SSIM | 0.4854 | 0.9464 | 0.8186 | 0.8016 | 0.8881 |
|
Table 1. Performance comparison between proposed denoising method and traditional filtering methods
Experimentaldevice | Parameter | Value |
---|
Light source | Centerwavelength /nm | 550 | CMOS detector | Cell size /mm | 0.00345 | Lens | Focal length /mm | 50 | | Posteriorintercept/mm | 12.4 |
|
Table 2. Parameters of experimental devices
Parameter | Noise image | Mean filteringmethod | Mediumfiltering method | Wavelet filteringmethod | Proposedmethod |
---|
MTF curve area | 0.0563 | 0.0547 | 0.0706 | 0.0572 | 0.1403 | PSNR | Inf | 36.2199 | 36.4075 | 36.8810 | 32.1785 | SSIM | 1 | 0.8415 | 0.8337 | 0.8318 | 0.8283 |
|
Table 3. Comparison between proposed denoising method and traditional filtering methods