<!doctype html public "-//w3c//dtd html 4.0 transitional//en"> <html> <head> <meta http-equiv="Content-Type" content="text/html; charset=iso-8859-1"> <meta name="Author" content="Mark Danks"> <meta name="Author" content="IOhannes m zm�lnig"> <title>Pixes (image processing)</title> </head> <body> <center> <h2> <u>Image processing</u></h2></center> The pix objects are used to do image processing to pixel data. If you load in an image with <i>[pix_image]</i>, then you can change what the image looks like before rendering it out <p>In general, processing images is <i>extremely</i> expensive, so you probably cannot have that many active pix objects. GEM only reprocesses images when the source image changes or one of the parameters for a pix object changes. This means that GEM will only process an image when something is different, instead of every frame. If you want to do a lot of processing at start up, but then not change anything once the patch is running, GEM will only do the computation once.<br> Modern CPUs use SIMD (Single Instruction - Multiple Data) (like MMX, SSE2, altivec) to make pixel-processing more effective (by processing data parallely). Until now, only the macOS version of Gem has support for SIMD for some pix-objects. MMX/SSE2 boosts will hopefully come in future Gem-releases. <p>The pix objects are divided into two general groups, those which take one input, and those which require two input images. For example, <i>[pix_invert]</i> will "invert" all of the pixels (if a pixel is white, it will change to black), while <i>[pix_add]</i> will add two images together. <p>Only some of the pix objects are described here. Look in the reference patches for explanations for the other pix objects. <p><a href="#invert">[pix_invert]</a> - invert the pixel data <br><a href="#add">[pix_add]</a> - add two pixes together <br><a href="#mask">[pix_mask]</a> - create an alpha mask <br><a href="#convolve">[pix_convolve]</a> - convolve a pix with a kernel <p><img SRC="tribar.gif" height=13 width=561> <h3> <a NAME="invert"></a>[pix_invert]</h3> <i>[pix_invert]</i> inverts the pixels in an image. To use <i>[pix_invert]</i>, simply make sure that you have already loaded an image into the chain. In the following patch, the fractal image will be inverted. <center> <p><img SRC="invert.jpg" BORDER=1 height=120 width=179></center> <p>Here is the difference between the fractal image and the inverted version. <center> <p><img SRC="normalFrac.jpg" height=256 width=256><img SRC="invertFrac.jpg" height=256 width=256></center> <p><img SRC="tribar.gif" height=13 width=561> <h3> <a NAME="add"></a>pix_add</h3> <i>[pix_add]</i> does what you would expect. It adds two images together. <center> <p><img SRC="add.jpg" BORDER=1 height=152 width=305></center> <p>This patch adds the fractal image with a car image. The processed image will often contain a lot of white pixels, because the data is just added together. This occurs in the resulting image, shown below. <center> <p><img SRC="addResult.jpg" height=257 width=255></center> <p><br> <p><img SRC="tribar.gif" height=13 width=561> <h3> <a NAME="mask"></a>pix_mask</h3> <i>[pix_mask]</i> is used to create an alpha mask from another image. In the following example (gem_pix/gemMaskDancer.pd), the fractal image's alpha channel is replaced by the dancer image. If the <i>[alpha]</i> object was removed, then you would just see the solid fractal image (because the alpha channel wouldn't be used). <p>In other words, images are composed of a red, a green, a blue, and an alpha channel. The alpha channel is the transparency of the pixel. <i>[pix_mask]</i> only modifies the alpha channel and does not touch the red, green, or blue data. <center> <p><img SRC="mask.jpg" BORDER=1 height=262 width=191></center> <p>The result is this image. <center> <p><img SRC="maskResult.jpg" height=218 width=187></center> <p><img SRC="tribar.gif" height=13 width=561> <h3> <a NAME="convolve"></a>pix_convolve</h3> <i>[pix_convolve]</i> convolves pix data with a convolution kernel. Basically, you can get really nice effects if you choose the correct kernel...and garbage if you choose the wrong one. <p>Edge detection is done with a convolution kernel, as is smoothing. The biggest problem is that convolving an image is about the most expensive operation that you can do in GEM. <p>Look at gem_pix/gemPixConvolve.pd to get an idea of some of the kernels that you can send to <i>[pix_convolve]</i> and the effects that you can get. <p>If you want to learn the math behind convolution, then find any standard image processing (or audio processing book, this is just 2D convolution). <br> <p><img SRC="tribar.gif" height=13 width=561> <p><a href="index.html">[return]</a> <br> </body> </html>