好的,
ICM 是Intersecting Cortical Model的缩写,是PCNN网络的简化。由于ICM模型本身是对哺乳动物视觉神经模型的模拟,所以在image processing方面与其他的模型相比有比较大的优势。
模型
In the ICM the state oscillators of all the
neurons are represented by a 2D array F (the
internal neuron states; initially Fij=0 8i, j) and the
threshold oscillators of all the neurons by a 2D
array Y (initially Yij=0 8i, j). Thus, the ijth
neuron has state Fij and threshold Yij. They are
computed from
Fij[n+1] = fFij[n] + Sij +Wij{Y[n]} (1)
Yij[n+1] = 1, if Fij[n+1]>Tij[n] (2)
0, otherwise
Tij[n+1] = gTij[n] + hYij[n+1] (3)
where Sij is the stimulus (the input image, scaled so
that the largest pixel value is 1.0); Yij is the firing
state of the neuron (Y is the output image); f ; g;
and h are scalars (examples of values are 0.9, 0.8,
and 20.0, respectively); Wij is the connection
function through which the neurons communicate;
in the first test below it modifies the neural
connections as centripetal autowaves according
to a curvature flow model; pulse segments in
iteration n will modify the neural connections such
that neurons near the edge of the segment will
increase their connection strength towards the
local centre of curvature; in the second test below
it is just a smoothing function; the smoothing of
image pixels will affect neighbouring pixels, i.e.
creating a connection between them; the actual
choice of smoothing function is not crucial; and
n=1,y,N is the iteration number. The scalars f
and g are decay constants and thus less than 1. In
order to ensure that Fij eventually becomes Yij, we
have f > g: The firings computed from Eq. (2) are
the output images of the ICM. Hence, the images
are called pulse images.
需要相关论文的可以发站内信给我,我也是做毕设的时候才用到的这个