JAI: GaussianBlur problem

Question

I have made a GaussianBlur class to control image Gauss blur .
but the larger radius I give the kernel, the thicker edge the image gets.
what's the problem?
// to generate gauss kernel
private static float[] makeKernel(double radius) {
radius += 1;
int s = (int)radius;
int size = s*s;
float[] kernel = new float[size];
double v;
for (int i=0; i<s; i++) {
     for ( int j= 0; j < s; j++) {
     kernel[i*s + j] = (float)(Math.exp(-0.5*(sqr(i)+sqr(j))/sqr(radius))/(2*Math.PI*sqr(radius)));
     return kernel;
public void applyFilter(float radius) {
     float[] elements = makeKernel(radius);
     normalize(elements);
     KernelJAI kernel = new KernelJAI((int)radius+1, (int)radius+1, elements);
     // start convolve image on the kernel
     ParameterBlock pb = new ParameterBlock();
pb.addSource(source);
pb.add(kernel);
RenderedImage target = JAI.create("convolve", pb, null);

Answer 1

the problem is how you want to treat to border of your image. The blur works by calculating the average value of the neighbouring pixels for a given center pixel. So you have to tell your algorithm what to do if your "center" pixel sits on the outside border and lacks its neighbours...
take a look at the docu on java.awt.image.ConvolveOp which specifies two types of "edge_conditions", i.e. solutions to the problem: EDGE_NO_OP, EDGE_ZERO_FILL