Identification of tobacco disease using content-based image retrieval with interactive image segmentation

Traditional tobacco disease recognition depends on tobacco experts or some complicated biological or chemical experiments, which is inconvenient and time consuming. Here is an image retrieval method introduced to identify tobacco disease type by means of a self-designed software system. Graph-based image segmentation and interactive segmentation methods are applied to extract the useful information of tobacco leaves infected with disease, and then specific descriptors are used to compute similarity or dissimilarity between these objects as the preparation for image retrieval. The whole software is written by C# language, and is intended for tablet PC. In the experiment, the software has a good performance for image retrieval of tobacco diseases and we have also confirmed that image retrieval can be applied to tobacco disease recognition as well. © 2012 Praise Worthy Prize S.r.l.