The following code does not assume that every grid location will have a value given, and also it does not assume that the coordinates are equally spaced. If there are multiple values for the same location, then it will average the values.
If you have a full grid of values, and the x are in strictly increasing order, and the distance between coordinates is consistent for each dimension, then the solution @KSSV shows will work just fine and with lower cost. This present code is for the case where those constraints do not hold -- this code handles scattered coordinates.
Note: this code does not assume that x and y are to the same scale. If, for example, your x is 0, 0.5, 1, 1.5, 2, and your y is 0, 1, 2, then the code will figure that you have different x and y resolutions, and will not decide that your real intention was to have implied 0's at y = 0.5, y = 1.5
data = [0 0 255
0 0.5 70
0 1 111
0.5 0 26
0.5 0.5 26
0.5 1 255
1 0 108
1 0.5 26
1 1 70];
x = data(:,1);
y = data(:,2);
z = data(:,3);
ux = unique(x); dx = min(diff(ux));
uy = unique(y); dy = min(diff(uy));
xidx = 1 + floor((x - ux(1))/dx);
yidx = 1 + floor((y - uy(1))/dy);
img = uint8(accumarray([yidx,xidx], z, [], @mean))
imshow(img)
