where f i and g i are two pixel-wise operators.
Moreover, since both f i and g i are pixel-wise operators, the computational complexity of GAC is the same as that of the AC model, i.e., O(C n 2log2 n), where n 2 is the number of pixels of the image x.
○, are element-wise operators.
We have R = R. / diag R diag R T, where./ and are element-wise operators.
Such pixel-wise operator inherently takes advantage of massively parallel architectures like GPU and FPGA.
where ( W a T ) k a × n is the PCA (whitening) transform, (PSDB X k))n×1 is the power spectral density vector of frame k, ∂ PSD B X ( k ) ∂B n × M is Jacobian of PSD vector with respect to B and ⊘ is the element-wise operator which divides each column of the left matrix by the right vector.
Then, the measurement matrix Φ ′, composed of L different block-wise sampling operators, is adopted to these decompositions with one operator for each level.
where k i is the i-th basis convolution kernel, α i denotes the i-th weight matrix, ⊗ denotes the convolution operation, ∘ denotes the pixel-wise multiplication operator, and C is the number of basis convolution kernels.
where ⊙ is the element-wise product operator and the fraction is the element-wise division operator.
where ⊕ denotes the bit-wise XOR operator.
where λ is a forgetting factor and ⊘ is an element-wise division operator.
