Double Variable Precision Rough Set Model And Its Application In Scene Image Segmentation

The advent of the big data era directly led to the rapid development of databasetechnology and computer data processing technology, which makes people have the ability toprocess the complex, diverse, disorderly mass data. Thus we can predict the future trend ofdevelopment from the potentially valuable rules or relationship of these data. As a kind ofsoft computing mathematical tool dealing with uncertainty, Rough set plays an important rolein the field of data mining.Focused on the underdeveloped robustness when the classical rough set modelencounters the noise for the incomplete information system, the expanded research for the“graining” and the “approximation” of the classical rough set model was studied, and the newgraining-approximation model was given, and its application in the scene image segmentationwas constructed. Specifically, the relationship between the Pawlak’s rough set model, theZiarko’s variable precision rough set model, and the Yao’s decision-theoretic rough set modelwas described with the help of the “rough membership degree”. The backward greedyattribute reduction algorithm and the forward greedy attribute reduction algorithm based onclassic rough set were depicted, by the analysis it is concluded that the latter algorithm isbetter. The attribute reduction algorithm based on the Ziarko’s variable precision rough setmodel was proposed on the basis of the concept “approximate attribute reduction” poweredby Beynon M. The necessity of adjusting the size of basic knowledge granule as well asintroducing the relative degree of misclassification was analyzed. Then the variable precisionrough set model based on variable-precision tolerance relation,(short for VPRS-VPTR) wasestablished on the basis of the object connection weight matrix, which was proposedaccording to the lack probability of system attribute value. Moreover the properties of theVPRS-VPTR model were discussed, the classification accuracy under the basic knowledgegranule size and the relative degree of misclassification were analyzed, the corresponding algorithm was depicted and the time complexity analysis was given afterwards. Theexperimental simulation results show that the classify accuracy of the VPRS-VPTR model ishigher when compared with others′research about the expansion rough sets, and thechange trends of the classify accuracy are similar for the trains and the tests of the severalgroups incomplete data sets on UCI database. Thus, the proposed model is more precise andflexibility, and the algorithm is feasible and effective. New heuristic rules were giving for theattribute reduction of the VPRS-VPTR model: the classification ability of the system is notreduced just as the attribute dependence is not less; the system recognition accuracy is as highas possible just as the similarity of the system’s domain is as large as possible.The VPRS-VPTR model was applied to the scene image segmentation, the doublevariable precision rough set model of the image was established, the image segmentationalgorithm combining with the rough entropy was given, experiments proves the algorithmis feasible.