Fig 1.
Some examples of motion blur in traffic sign images with different sizes.
Fig 2.
Exemplar images and the corresponding contour masks.
Table 1.
Matching calculation cost comparison.
Table 2.
Matching accuracy comparison.
Fig 3.
(a)—(f) denote sharp image, the corresponding blurred image, NCC result, MI, ECC and our method, respectively, about an example of speed limit 20 signs. (g)—(l) denote sharp image, the corresponding blurred image, NCC, MI,ECC and our method, respectively, about another example of give way signs.
Fig 4.
Kernel similarity comparison.
Fig 5.
(a)Real blur kernel (b) estimated kernel with L2-norm (c) estimated kernel with L0.5-norm (the true kernel is provided in [14]).
Fig 6.
Quantitative comparisons among blind deblurring methods.
Fig 7.
Image deblurring comparison for synthetic images.
(a) denotes the input image and real blur kernel. (b) denotes the matched exemplar. (c) denotes the predicted ∇S. (d)—(j) denote the deblurring results of Krishnan[12], Xu eta.[17], Shan[15], Xu & Jia[7], Levin[16], Pan[9] and our method, respectively. In (d)—(j), each image includes deblurring image, estimated kernel and error ratio.
Fig 8.
Deblurring example suffering from a heavier blur.
(a) denotes the input image and real blur kernel. (b) denotes the matched exemplar. (c) denotes the predicted ∇S. (d)—(j) denote the deblurring results of Krishnan[12], Xu eta.[17], Shan[15], Xu & Jia[7], Levin[16], Pan[9] and our method, respectively. In (d)—(j), each image includes deblurring image, estimated kernel and error ratio.
Fig 9.
Example of real captured image.
(a) denotes the real captured image. (b) denotes the matched exemplar. (c) denotes the predicted ∇S. (d)—(j) denote the deblurring results of Krishnan[12], Xu eta.[17], Shan[15], Xu & Jia[7], Levin[16], Pan[9] and our method, respectively. In (d)—(j), each image includes deblurring image and estimated kernel.
Fig 10.
Deblurring results for synthetic traffic signs during sunny day time.
(a) denotes the input image and real blur kernel. (b) denotes the blurred image. (c) and (d) denote the deblurring results of Pan[9] and our method, respectively.
Fig 11.
Deblurring results for synthetic traffic signs during cloudy day time.
(a) denotes the input image and real blur kernel l. (b) denotes the blurred image. (c) and (d) denote the deblurring results of Pan[9] and our method, respectively.
Fig 12.
Deblurring results for synthetic traffic signs during foggy day time.
(a) denotes the input image and real blur kernel. (b) denotes the blurred image. (c) and (d) denote the deblurring results of Pan[9] and our method, respectively.
Fig 13.
Deblurring results for synthetic traffic signs during late morning.
(a) denotes the input image and real blur kernel. (b) denotes the blurred image. (c) and (d) denote the deblurring results of Pan[9] and our method, respectively.
Fig 14.
Deblurring results for synthetic traffic signs during early evening.
(a) denotes the input image and real blur kernel. (b) denotes the blurred image. (c) and (d) denote the deblurring results of Pan[9] and our method, respectively.