The Kohonen software is designed to provide image color mapping capabilities, allowing users to efficiently assign colors to individual pixels. With its intuitive interface and powerful algorithms, Kohonen is a valuable tool for artists, designers, and other professionals seeking precise color mapping solutions.
SOM is known as one of the most prevalent neural computation methods in use, and several thousand scientific articles attest to its popularity. One of the major advantages of SOM is its ability to produce visualizations of high-dimensional data, making it especially useful for data-heavy projects.
One of Kohonen's standout features is its ability to be used as space filters when considering an image. A Kohonen map is capable of finding the most significant filters of a given size that can correctly represent the image. The process involves considering input vectors as sub-images of a specific size (for instance, 5x5 pixels) and feeding the network with this data. Once the learning process is complete, the filters can be used to reconstruct the entire image successfully.
Another way to use Kohonen in tandem with an image is as color filters. In this application, a Kohonen map is used to find statistically significant colors in the image. Here, an input vector is a colored pixel from the image (with three dimensions: red, green & blue). After learning is complete, it is possible to use prototypes that have been identified to get a fair reconstruction of the image.
Kohonen usage is quite simple, with the user providing the image file, and changing the "-size" parameter to alter the size of space filters. Additionally, "-epochs" and "-lrate" can be modified to adjust learning time and the learning rate, respectively. The "-mode" parameter is used to determine if Kohonen map will be used as color or filter filters, and we can change the "-seed" parameter to specify a random seed. Finally, the "-zoom" parameter alters image display zoom parameters.
Version 1.0: N/A