Using a Higher Precision (than 8-bit unsigned integer) Buffered Image for Heightmaps in Java

Question

I am generating a heightmap for every quad in my quadtree in openCL. The way I was creating the image is as follows:

DataBufferInt dataBuffer = (DataBufferInt)img.getRaster().getDataBuffer(); int data[] = dataBuffer.getData(); //img is a bufferedimage inputImageMem = CL.clCreateImage2D( context, CL_MEM_READ_WRITE | CL_MEM_USE_HOST_PTR, new cl_image_format[]{imageFormat}, size, size, size * Sizeof.cl_uint, Pointer.to(data), null);

This works ok but the major issue is that as the quads get smaller and smaller the 8-bit format of the buffered image starts to cause intolerable "stepping" issues as seen below:

enter image description here

I was wondering if there was an alternate way I could go about doing this?

Thanks for the time.

Is this the "stepping" you're talking about? i.imgur.com/Sx9egmm.png?1 — House
– House, Commented Aug 11, 2013 at 0:20
Can you change the title to something more related to your question? I was trying to come up with something but didn't want to misinterpret your question. — House
– House, Commented Aug 11, 2013 at 5:10
Why do you have to use a BufferedImage? What format are your input images? — msell
– msell, Commented Aug 12, 2013 at 4:58

Adam · Accepted Answer · 2013-08-11 21:41:04Z

1

Are you using texture filtering when reading the height map? I'd expect that a bilinear filter would smooth things out a bit.

If you still need better precision then change the image format to something like CL_UNORM_INT16 or CL_FLOAT - see http://www.khronos.org/registry/cl/sdk/1.0/docs/man/xhtml/cl_image_format.html

I'm not quite sure how you do that from Java though.

answered Aug 11, 2013 at 21:41

Adam

7,72921 silver badges26 bronze badges

\$\begingroup\$ Yea that's the issue: I can't figure out how to use anything other than 8-bit integer with a BufferedImage. \$\endgroup\$

multifractal
– multifractal

2013-08-11 21:48:52 +00:00
Commented Aug 11, 2013 at 21:48
\$\begingroup\$ Since Java doesn't have native OpenCL support, it might help to know which library you're using. The other option would be to use a double width image, and decode it on the GPU as say (pixel[x]/256.0 + pixel[x+1]). \$\endgroup\$

Adam
– Adam

2013-08-12 19:51:17 +00:00
Commented Aug 12, 2013 at 19:51
\$\begingroup\$ I tried using bilinear interpolation with no luck... I am just using Java OpenCL (JOCL) \$\endgroup\$

multifractal
– multifractal

2013-08-13 02:08:42 +00:00
Commented Aug 13, 2013 at 2:08
\$\begingroup\$ The Java version of OpenCL seems to support these formats as well. cl_image_format has a image_channel_data_type member that can be used to specify the data format (eg. CL_FLOAT etc.) \$\endgroup\$

bummzack
– bummzack

2013-08-13 07:37:20 +00:00
Commented Aug 13, 2013 at 7:37
2

\$\begingroup\$ docs.oracle.com/javase/7/docs/api/java/awt/image/… says BufferedImage supports TYPE_USHORT_GREY which is 16-bit. \$\endgroup\$

Adam
– Adam

2013-08-13 08:31:27 +00:00
Commented Aug 13, 2013 at 8:31

Add a comment |

Stack Exchange Network

Using a Higher Precision (than 8-bit unsigned integer) Buffered Image for Heightmaps in Java

1 Answer 1

You must log in to answer this question.

Hot Network Questions

Using a Higher Precision (than 8-bit unsigned integer) Buffered Image for Heightmaps in Java

1 Answer 1

You must log in to answer this question.

Related

Hot Network Questions