Recognition System Based On Rough Set Theory, Digital Image Processing

A system supporting image processing and classification is built in this thesis. The system is programmed MATLAB Language. Image format conversion, image adjusting, image filtering, FFT and image edging detection are implemented in the part of image processing.The system supporting image classification is constructed based on the Rough Sets approach includes training step and classification step. Under training step, each raw texture image is preprocessed, image subdivides, feature extracted, feature zoning, added into the Rough Set training database namely Raw decision table. In the decision table, condition attributes correspond to the extracted feature and decision attributes to classification value of each sub-images namely the number of Raw texture image. Through implementation of the Rough Set approach, under the conception of indiscernibility relation, the meaningful features are collected, the decision table is redacted and the minimum decision rules are deduced. Under image classification step, testing database is built in the way similar to the one under the training step. By using the classifier obtained under the training step, image classification is finished, and the result is outputted. Through utilization of the rough set approach, the number of image features is significantly reduced, the structure of system reduced, and then the running speed improved.Experimentally, under image classification, about 85.00% classification results can be obtained. However, after injection noise into image, the results fall dramatically to 48%.