In this post i would like to demonstrate the ability of R to handle and classify image data with the help of ImageJ and Rserve bundled and implemented in Bio7. In general R is a very useful application for image analysis and plenty of “pure” R packages for image analysis are already available. But the combination of a scientific image analysis tool like ImageJ with R offers much more possibilities to edit and analyze images. With the correct image datatype (raw, integer, double) R is also able to handle and process images >2000*2000 pixels (in combination with java!).
For my own curiosity i tried to find out the limits on a 32-bit and 64-bit OS with the image data present in the virtual maschine and after transfer (by means of Rserve) in R, too.
Hardware: AMD Turion 2.0 GHz Dual core, 3Gb RAM Windows Vista 32-bit
Java enabled memory: 1024Mb
Cluster Analysis (clara): Max 5000*5000*3(RGB) -> 6 centers: 70sec., 12 centers: 100s (byte transfer!).
Transferred 7 Landsat images 7457*6991 (byte transfer) and created a new image from the sum of the 7 images (integer conversion! with a triggering of the garbage collector).
Transferred 8000*8000 image as integer data to R in one shot.
Transferred and returned a 16000*16000 image as byte data (grayscale) to R and ImageJ in one shot.
And with 64-bit (with a personal OSX port of Bio7).
Mac OSX 10.6, 2.93 GHz, 4GB RAM
Java enabled memory: 2512Mb
Cluster Analysis (clara) with R: Max 7000*6157*3(RGB) -> 6 centers: 168sec. (byte transfer!).
R with ImageJ can be used perfectly to classify images. Two examples are shown below. The first example uses the clara algorithm which is generally more appropriate for images >2000*2000.
Example 1: Unsupervised classification with the Bio7 interface (clara).
Example 2: Supervised classification (rpart) with the Bio7 interface.
This examples also demonstrate the general use of R for image analysis. Some more examples of image analysis with R, ImageJ and the Bio7 interface can be found here: