Image acquisition using oversampled one-bit poisson statistics

An image acquisition method using an image sensor comprising an array of one-bit pixels,

• wherein the size of the pixels is below the fundamental diffraction limit of visible light, so that the image sensor is oversampling the optical resolution of visible light,
• wherein each pixel in said image sensor delivers a one-hit quantized measurement of the local light intensity received at that pixel during an acquisition period,
• wherein an image is reconstructed from the measurements of said pixels using a maximum likelihood estimation algorithm in order to retrieve coefficients (cn) of a sum of interpolation kernels corresponding to said image,
• wherein the numerical solution of said maximum likelihood estimator is computed using an iterative method.

A method of operating an image sensor having an array of binary pixels, the method comprising:

• during a first frame interval, sampling pixels in the array of binary pixels with a first oversampling factor;
• during a second frame interval substantially equivalent in exposure value to the first frame interval, sampling pixels in the array of binary pixels with a second oversampling factor, the second oversampling factor exceeding the first oversampling factor;
• and
• creating a multilevel-pixel downsampled image from pixel samples gathered during the second frame interval, wherein:
• the ratio of pixel samples gathered during the second frame interval multilevel pixels is substantially equal to the second oversampling factor,
• multilevel-pixels in the downsampled image have a value corresponding to a maximum-likelihood estimate of the light exposure for the pixel samples corresponding to that multilevel pixel, and
• the maximum likelihood estimate for a multilevel sample is
• c
• ^
• ML
• ⁡
• (
• b
• )
• =
• {
• Kp
• 0
• [
• -
• 1
• ]
• ⁡
• (
• 1
• -
• K
• 1
• /
• K
• )
• ,
• if
• ⁢
• ⁢
• 0
• ≤
• K
• 1
• ≤
• K
• ⁡
• (
• 1
• -
• p
• 0
• ⁡
• (
• S
• /
• K
• )
• )
• ,
• S
• ,
• otherwise
• ,
• where K1 is the number of pixel samples, corresponding to the multilevel sample, that detected a threshold number of photon strikes, p0(−1)(x) is the inverse function of p0(x), p0(x) expresses the likelihood of K1 given a light intensity c, S is a maximum result value, and K is the second oversampling factor.