Image Classification Based On The Tolerance Information Granules Space

The mode of tolerance information granular space is a kind of granular computing models which is constructed based on the tolerance relations. This model has advantages in granulation of existing knowledge, resources and information at different levels and size, and building the relations among granules. By discovering the useful information contained by the knowledge and resources, the model can provide a better insight into their nature.On the basis of studying fundamental principles and techniques of granular computing, this paper analyzes tolerance information granular space and proposes an algorithm for rales mining based on the tolerance information granular space. Then it is applied in image pattern recognition. Against the background of image, the paper stresses on the construction, management and computing of character tolerance information granules. What's more, a classifier based on image character tolerance information granular is advanced in the paper by analyzing and studying the image character tolerance information granular space of different granularities of characteristic tolerance information granule. Besides, simulation classification tests are done towards five different image classes. It's proved that the classifier based on tolerance information granule is effective. Finally, in order to increase the accuracy of classification, the above algorithm is improved by using image spatial distribution information, so that a hierarchy image classifier based on global-local image features is composed with the presentation of a corresponding demonstration system.The last part of the paper, a summary is given. It analyzes areas for improvement and looks forward to the further study direction of information granule and granular computing technology in the field of image classification.