Research On Data Mining Methods Based On Fuzzy Sets And Decision-Theoretic Rough Sets And The Application In Image Segmentation

Fuzzy sets and rough sets are effective tools to solve imprecision and uncertainty in data mining.When huge amount of data is generated by people,the data itself is characteristic of uncertainty.Meanwhile,problems which are to be solved may contain some fuzzy concepts.Therefore,it is necessary to make some research on data mining methods based on fuzzy sets and rough sets.As is known to all,rough sets model needs to process continuous data for neces-sary discretization,which must cause information loss to a certain extent.The fuzzy rough sets theory which combines the fuzzy sets theory with rough sets model can better deal with continuous data.The decision-theoretic rough sets model(DTRS)that has extended the traditional rough sets model is now widely used.How to combine fuzzy sets theory with decision-theoretic rough sets model in attribute reduction and classification for continuous data is worth studying.As an aspect of unsupervised learning in fuzzy theory,fuzzy clustering analysis has also been widely concerned,especially Fuzzy C-Means(FCM)clustering algorithm for image segmentation.Fuzzy C-Means(FCM)is an unsupervised clustering method for iterative optimization based on the objective function.Soft partition using FCM can practically reflect the fuzziness and uncertainty of images,therefore,its performance is superior to the traditional hard clustering method.However,there still exist some problems when the FCM algorithm is utilized in image segmentation.On one hand,it is difficult to determine the number of clusters and cluster centroids.Poor initialization has a negative effect on the image segmentation result.On the other hand,the FCMalgorithm does not make use of space information of adjacent image pixels,which may result in incomplete segmentation model.This thesis respectively focuses on the above-mentioned issues and makes im-provements.The main work is as follows:Fuzzy membership function is employed to change the method of calculating the conditional probability contained in the risk loss functions in the DTRS mod-el.Thus,we can obtain the fuzzy decision-theoretic rough set model(FDTRS).Based on this new model,we can conduct fuzzification and reduction for at-tributes and classify objects according to new rules.Experiments show that the method is effective.Before utilizing FCM to segment images,we propose the method which em-ploys the extended decision-theoretic rough set model(DTRS)for clustering va-lidity analysis.Thus,it can determine the optimal number of clusters and cluster centroids,which will avoid blind initialization in FCM algorithm.Experimental results have proved the effectiveness and more advantages compared with other methods such as colony FCM algorithm.We propose an image segmentation method which combines the FCM clustering and graph cut theory.We presegment the image into a set of superpixels and then use FCM algorithm for image segmentation.According to the clustering result-s and the relationship between adjacent superpixels,we can construct a graph model and generate weights of edges.Therefore,we achieve the goal of making the most of image space information for subsequent segmentation.Experimental results show that this method can achieve good segmentation results on partial UC-Berkeley image dataset.