The result is shown in figure 1. A Christmas stocking was filmed with a blue background with a Sony
mini DV DCR HC90 camera, extracted and composited with a different background. The different background was captured using a PixelLink PL A7-42 camera and with the Sony camera for the reference shot.
The distance from the fitted line threshold was set to
5.0 times the standard deviation, the minimum size for
regions was set to 200 pixels, k-means was performed
with 10 classes and the tolerated proportion of inconsistent pixels was set to 5%. Five iterations were performed on each of the 20 images (720 × 480 pixels)
contained in the video sequence. Convergence rarely
exceeded five iterations for images of this size.
For figures 1(e) and 1(f), the grey level corresponds
to the local level of noise. We see that parts of the stocking has brighter pixels in figure 1(e), the alteration was
detected. However, the darker region of the stocking
does not stand out from the background. Indeed, c was
computed to be 0.010 for the PixelLink camera versus
0.015 for the Sony camera, while the computed values
for d were 0.77 and 0.70 respectively. The two fitted
lines thus intersect at grey level 14. As a consequence,
the noise variance is very similar for both cameras for
dark areas. The thresholded result of figure 1(g) confirms the detection of the alteration.
The right column of figure 1 shows the results for the
reference video. All parameters were the same. Since
The result is shown in figure 1. A Christmas stocking was filmed with a blue background with a Sony
mini DV DCR HC90 camera, extracted and composited with a different background. The different background was captured using a PixelLink PL A7-42 camera and with the Sony camera for the reference shot.
The distance from the fitted line threshold was set to
5.0 times the standard deviation, the minimum size for
regions was set to 200 pixels, k-means was performed
with 10 classes and the tolerated proportion of inconsistent pixels was set to 5%. Five iterations were performed on each of the 20 images (720 × 480 pixels)
contained in the video sequence. Convergence rarely
exceeded five iterations for images of this size.
For figures 1(e) and 1(f), the grey level corresponds
to the local level of noise. We see that parts of the stocking has brighter pixels in figure 1(e), the alteration was
detected. However, the darker region of the stocking
does not stand out from the background. Indeed, c was
computed to be 0.010 for the PixelLink camera versus
0.015 for the Sony camera, while the computed values
for d were 0.77 and 0.70 respectively. The two fitted
lines thus intersect at grey level 14. As a consequence,
the noise variance is very similar for both cameras for
dark areas. The thresholded result of figure 1(g) confirms the detection of the alteration.
The right column of figure 1 shows the results for the
reference video. All parameters were the same. Since
การแปล กรุณารอสักครู่..
