Tolerance Neighborhood Model And Its Application In The Scene Images With Occluded Objects

With the rapid development of the large data processing technology,People's life habits and the way of working and thinking model have been greatly changed.Many people have realized the vast prospects of the huge amounts of data analysis and processing,which has moved them with a hope of getting useful information as far as possible.However,in terms of the characteristics of big data,such as imprecise,inconsistent,high dimension and incomplete information and knowledge,the rough set has been widely used in big data mining as an effective tool to deal with uncertainty problem.Current neighborhood rough set has been usually used to solve complete information systems,not incomplete systems.In this paper,we propose a new neighborhood model to extend the classical rough set theory by using the concept of "neighborhood" and "tolerance complete degree",and apply the model to image classification of occluded objects.Specifically,we introduce the neighborhood rough set model based on the Pawlak's classical rough set,in case of dealing with the mixed data types at the same time,in which nominal attributes and numeric attributes included.And the forward greedy attribute reduction algorithm was proposed based on neighborhood rough set.By analyzing the algorithm result,the attribute reduction algorithm without tolerance capacity cannot get the ideal result in dealing with the incomplete information systems.On the basis of the theory,the extend tolerance relation model is proposed.The degree of complete tolerance and neighborhood threshold is used as the constraint conditions to find the extended tolerance neighborhood.Based on this neighborhood,the decision positive region and the attribute importance is achieved,and get the attribute reduction algorithm base on extended tolerance relation by the importance which regarded as a heuristic factor.By removing redundant attribute to reduce the impact of noise data on classification results.Analysis of the change of the classification accuracy trend for the single sample set in different classification algorithm and different neighborhood threshold.The proposed method use 7 groups of different type of data sets on UCI database for simulation,and compare with other rough set method respectively.By analyzing classification accuracy trend of all the algorithm,we can come up with the result: ETR algorithm can keep a good reduction,meanwhile it does not reduce the classification accuracy.So the effectiveness and feasibility of the algorithm for the extend tolerance neighborhood model is verified in dealing with the incomplete information system.In this paper,the extended tolerance relation is applied to the scene image classification of target occlusion based on the method of color and texture fusion.Firstly,the knowledge representation system and the object-oriented set system of the occlusion image object set are constructed.The extended tolerance neighborhood model is used to establish the tolerance granularity space of the image edge and occlusion boundary.Secondly,the statistical characteristics of the histogram is obtained by calculating the histogram of the color feature in the tolerance granularity space.Finally,under the various contrast algorithms,the proposed algorithm is validated by different classifiers.The experimental results show that the method can solve the problem of occlusion in complex scene images and realize the classification and retrieval of scene images.