Author Affiliations
School of Electronics and Information Engineering, Lanzhou Jiaotong University, Lanzhou, Gansu 730070, Chinashow less
Fig. 1. Schematic diagram of Criminisi algorithm
Fig. 2. Changes in various data during mural restoration. (a) Mask image; (b) change in confidence; (c) change in date item; (d) change in priority
Fig. 3. Restoration effect of images at each threshold. (a)Original image; (b)mask image; (c) Yth=10000; (d) Yth=5000; (e) Yth=2000; (f) Yth=1000; (g) Yth=500; (h) Yth=200; (i) Yth=50; (j) dynamic threshold of proposed method
Fig. 4. Restoration results of mural 1. (a) Original image; (b) mask image; (c) traditional Criminisi method; (d) method in Ref.[9]; (e) method in Ref.[11] ; (f) proposed method; (g) enlarged restored area of traditional Criminisi method; (h) enlarged restored area of method in Ref.[9]; (i) enlarged restored area of method in Ref.[11];(j) enlarged restored area of proposed method
Fig. 5. Restoration results of mural 2. (a) Original image; (b) mask image; (c) Criminisi method; (d) method in Ref.[9];(e) method in Ref.[11]; (f) proposed method
Fig. 6. Restoration results of mural 3. (a) Original image; (b) mask image; (c) Criminisi method; (d) method in Ref.[9];(e) method in Ref.[11]; (f) proposed method
Fig. 7. Restoration results of mural 4. (a) Original image; (b) mask image; (c) traditional Criminisi method; (d) method in Ref.[9]; (e) method in Ref.[11]; (f) proposed method; (g) enlarged restored area of traditional Criminisi method; (h) enlarged restored area of method in Ref.[9]; (i) enlarged restored area of method in Ref.[11]; (j) enlarged restored area of proposed method
Fig. 8. Mural restoration results in large areas. (a) Original image; (b) mask image; (c) Criminisi method; (d) method in Ref.[9]; (e) method in Ref.[11]; (f) proposed method
Fig. 9. Mural restoration results in real damaged areas. (a) Original images; (b) mask images; (c) Criminisi method; (d) method in Ref.[9]; (e) method in Ref.[11]; (f) proposed method
Method | SSDA with fixed threshold | Proposed method |
---|
10000 | 5000 | 2000 | 1000 | 500 | 200 | 50 |
---|
PSNR /dB | 32.56 | 33.71 | 32.82 | 33.76 | 33.69 | 32.20 | 32.09 | 36.68 | Time /s | 8.65 | 8.28 | 8.04 | 7.19 | 6.90 | 6.37 | 5.80 | 6.26 |
|
Table 1. PSNR and time under different methods
No. | Mask image | Criminisi method | Method in Ref.[9] | Method in Ref.[11] | Proposed method |
---|
1 | 22.1686 | 32.8772 | 35.2642 | 35.9254 | 37.1888 | 2 | 24.6271 | 33.0012 | 34.1628 | 33.0956 | 35.0714 | 3 | 26.5476 | 37.1448 | 40.4067 | 39.2145 | 41.9898 | 4 | 34.3625 | 40.3273 | 43.4216 | 41.8620 | 43.5608 | 5 | 30.2568 | 37.2985 | 39.9856 | 38.7564 | 40.0258 | 6 | 26.2545 | 30.1698 | 32.4183 | 32.4215 | 32.6986 | 7 | 25.6589 | 30.5896 | 31.5244 | 30.9896 | 31.8652 | 8 | 23.2546 | 29.3256 | 29.1589 | 29.3624 | 30.0526 | 9 | 32.2568 | 39.5428 | 41.0269 | 40.1563 | 42.0024 | 10 | 21.2351 | 33.4585 | 34.5624 | 34.8024 | 34.8569 |
|
Table 2. PSNR values of restoration results of each algorithmunit: dB
No. | Criminisi method | Method in Ref.[9] | Method in Ref.[11] | Proposed method |
---|
12 | 664.5629395.1932 | 557.4610346.5987 | 541.3256322.7564 | 376.2413284.8241 |
|
Table 3. Restoration time of each algorithm unit: s