Instead of using one photodiode in the illumination based Compressive Sensing Camera, what about using two or more single photodiodes at different locations so that they provide a 3D description of the Stanford Bunny being illuminated ? Since we are solving for A x_1 = b_1 for the first photodiode and A x_2 = b_2 for the second photodiode, then one can also solve for A x_1 = b_1 and A (x_2 - x_1) = (b_2 - b_1) and expect a much faster resolution for the second problem as x_2 - x_1 is likely to be very sparse as mentioned in Yin Zhang's presentation featured here.
Monday, July 28, 2008
CS: A 3D Illumination Compressive Sensing Camera System ?
Instead of using one photodiode in the illumination based Compressive Sensing Camera, what about using two or more single photodiodes at different locations so that they provide a 3D description of the Stanford Bunny being illuminated ? Since we are solving for A x_1 = b_1 for the first photodiode and A x_2 = b_2 for the second photodiode, then one can also solve for A x_1 = b_1 and A (x_2 - x_1) = (b_2 - b_1) and expect a much faster resolution for the second problem as x_2 - x_1 is likely to be very sparse as mentioned in Yin Zhang's presentation featured here.
Not sure to understand correctly your comment. If you want to recover the bunny 3-D shape, I'm not sure that the 3-D surface information contained (somewhere) in x_1 and x_2 will accept a linear model as the one proposed.
ReplyDeleteI think it could work only approximately, or exactly at order 0 for non-occluding point-like objects.
Another point : are you sure that the matrix A is the same for the two observations ? I'm affraid that if A has to contain some geometrical information, you would have A_1 x_1 = b_1 and A_2 x_2 = b_2, and you cannot realize the difference you want.
Sorry, I have perhaps not understood the correct meaning of your post.
Best,
Laurent