Reducing graph data without losing graph shape

I know this question is quite old but I had a problem almost similar.

To reduce the number of points to display without affecting the shape of the graph, We use the Ramer-Douglas-Peucker algoritm. The difference of shape between the uncompressed graph and the one with this algorithm is unnoticeable.

It seems to me that 1 in 200 is pretty serious data loss, and if those 200 values that should be represented with one value on the graph aren't close enough to be meaningfully substituted with an average, you have yourself a problem. If average isn't good enough, you must find a criterium to tell what data is more significant and should be included, and we can't help you with it because we don't know what kind of data it is, its statistical properties, or why any value would be more significant than the other. With those additional info, maybe a more specific answer could be given.

EDIT: After looking at the graph, it seems that you need both minimum and maximum in a given interval, because the dark blue area are values between those two, correct? Maybe you can take 100 values and make a graph from minimum, maximum, and average, so that every point in graph is made with 6 instead of 200 values, or something like that.

Another approach that might work is splitting the graph up into 200 point bins, and discard all but the maximum, minimum, and median points in each interval. Each of the three points in the interval gets plotted at its original location, so the locations of the extreme values won't change. Using the median instead of the mean will probably work better for your data set because the maxima are much more extreme than the minima, which would cause the filtered graph to shift upwards if you used the mean.