The Jasmine Vision System Builder
Our research into vision systems that learn involves uses genetic programming to evolve programs that solve the individual stages of a complete system from a set of generic vision components using training data. Although that process is not interactive, a graphical interface (GUI) is the most convenient way to capture the training data, monitor the progress of the evolution, and so on. So, inspired by Aladdin, part of the GENIE Hyperspectral GA Environment, we developed Jasmine as a front-end and controlling GUI for our GP 'engine'.
Jasmine is written in Java. It allows you to capture the training
data for, and monitor the evolution of, the main stages of a complete
vision system. Jasmine lets you can create as many classes as are
appropriate for your vision problem, and then "paint" them onto
images; the places that you have painted are used as the training
data, first for segmentation and subsequently for classification.
Jasmine is project-based, so you can work on a set of images, classes,
and overlays easily without needing to reload stuff all the time.
What does Jasmine stand for? Err . . . let's see . . . Jasmine — A SegMented Image Notation Environment? Answers on a postcard . . .
You can run Jasmine by clicking the button to the right. As long as
your web browser is able to run Java .jar files, it should
run directly; otherwise, download the jar-file and use it on your
system in the usual way.
To show you how to use Jasmine, we have the following tutorials, listed in order of increasing complexity:
Remember that this is research code and is not guaranteed to work properly. Use it at your own risk!