Shadow Removal
黄伟俊 13307130251
Result
Detection
Goal:
Ms:which specifies an area that is entirely inside the shadow, with all of its pixels belonging to a single surface (a region with roughly uniform color and texture).
Ml:which specifies an area entirely outside the shadow, with all of its pixels belonging to the same surface as the previous mask.
Floodfill
Before GaussianBlur
After GaussianBlur
M_s
M_l = {surface} - M_s
surface
Illumination Recovery
I(x,\lambda)=L(x,\lambda)R(x,\lambda)\quad (1)
L(x,\lambda)=L^d(x,\lambda)+L^a(x,\lambda)\quad (2)
I^{lit}(x,\lambda)=L^d(x,\lambda)R(x,\lambda)+L^a(x,\lambda)R(x,\lambda)\quad (3)
I^{lit}(x,\lambda)=L^d(x,\lambda)R(x,\lambda)+\frac{1}{a(x)}I^{shadow}(x,\lambda)\quad (5)
Use method in "The Shadow Meets the Mask: Pyramid-Based Shadow Removal"
I^{shadow}(x,\lambda)=a(x)L^a(x,\lambda)R(x,\lambda) \quad (4)
Illumination Recovery
I^{lit}_{k}(p)=\alpha_k(p)+\gamma(p)I_k^{shadow}(p)
\alpha_{k},k\in \{R,G,B\}
\alpha_k=\mu_k(L)-\gamma\mu_k(S),k\in \{R,G,B\}
\gamma=\frac{\sigma(L)}{\sigma(S)}
More results
references
Text
Shadow removal
By erichuang1994
