Author Affiliations
School of Optical and Electronic Information, Huazhong University of Science and Technology, Wuhan, Hubei 430074, Chinashow less
Fig. 1. Block diagram of defect detection algorithm
Fig. 2. Flow chart of image preprocessing and its effect
Fig. 3. Button defect-free image reconstruction and residual image with defects. (a) Original image; (b) defect-free image; (c) residual image
Fig. 4. Pixel value distribution of original and residual images. (a) Original image; (b) residual image
Fig. 5. (a) Residual image; (b) binary image
Fig. 6. Result diagrams of different defection methods. (a1)-(a3) Original images of three samples; (b1)-(b3) results by background substraction method; (c1)-(c3) results by linear regression method; (d1)-(d3) results by proposed method
Fig. 7. ROC curves by proposed method
Fig. 8. Effect of different parameter C on detection effect. (a) Original image; (b) detection effect at C=1; (c) detection effect at C=3; (d) detection effect at C=6
Fig. 9. Recognition rate of algorithm versus illumination
Sample No. | Number ofdefect-free buttons | Number ofdefective buttons | Size /pixel×pixel |
---|
1 | 200 | 110 | 256×256 | 2 | 132 | 128 | 152×152 | 3 | 132 | 88 | 169×169 |
|
Table 1. Detail information of test samples
Sample No. | Subtraction method | | Linear regression method | | Proposed method |
---|
| TPR | | TNR | R | TPR | TNR | R | TPR | TNR | R |
---|
1 | 0.805 | 0.855 | 0.823 | | 0.935 | 0.827 | 0.897 | | 1.000 | 0.955 | 0.976 | 2 | 0.731 | 0.594 | 0.64 | | 0.981 | 0.875 | 0.910 | | 0.981 | 0.906 | 0.944 | 3 | 0.909 | 0.705 | 0.83 | | 0.944 | 1.000 | 0.9652 | | 0.986 | 1.000 | 0.990 |
|
Table 2. Results by different methods
Lighting condition No. | 1 | 2 | 3 | 4 | 5 |
---|
Light source output value | 150 | 155 | 160 | 165 | 170 | Actual light intensity /lx | 5100 | 6400 | 7500 | 8600 | 9800 |
|
Table 3. Correspondence between light source scale and actual light intensity