Research On Theory Of Granular Computing And Its Application On Image Retrieval

Granular Computing as a new mathematical tool is a new emerging subject of artificial intelligence. It has three main theories: Granular Computing Theory based on Fuzzy Logic, Granular Computing based on Rough Set, Granular Computing Theory Based on Quotient Space. Based on it, we solve the problems by the use of appropriate coarse grained in a different world to discuss issues and find the answers. As the coarse-grained world is simpler than the original one, we are able to narrow the scope and speed up the process when we apply it. However, sometimes we may lose some useful information. Therefore, when the problem cannot be solved in the coarse-grained world, it may be appropriate to reduce the size to get a more detailed level of space, find the answer to the question that cannot be confirmed in previous hierarchy. Thus, problems of uncertainty can be resolved in the proper level. Nowadays the theory of granular computing is widely used in knowledge discovery, machine learning, semantic web services, image classification and retrieval and so on, but it is still hard to apply it to practice.On the one hand, image retrieval calculates and compares the bottom features, such as color, texture, contour and shape to get the images collection which meets users'needs. On the other hand, we take semantic features into account. However, we have not got an effective way to improve the time complexity, recall rate and precision rate in the traditional image retrieval methods. In this article we study image retrieval techniques and theoretical methods of granular computing, trying to raise new image retrieval methods by combining the two above. The main innovation points of this paper are as follows:(1) By introducing the probability-rough set theory and image semantic annotation technology to the image information retrieval, this paper puts out a kind of image information retrieval model based on Na?ve Bayesian and probability rough set model. Firstly, we researched the crisp annotation and vague annotation of images, built crisp annotation space and put the crisp annotation images and the vague images to query. Secondly, we calculated the semantic similarity between query characteristics and the characteristics of image library, and output the retrieval results by ordering according to similarity.(2) For the problem of traditional methods for texture image retrieval in accuracy, a method was proposed which called multi-level image texture recognition based on tolerance granular space. Secondly, considering the loss of color in texture image, we addressed color matching method based on granularity level. Finally, Combining the above two methods, the paper gave the improved similarity algorithm which integrates color and texture features.(3) Basing on the granular computing theory, this paper builds rough granular theory, constructs rough granular model and similarity calculation method by introducing the concept of granular edges and layered entropy, then draws a method of image texture recognition. The method which this paper proposed increases the practical of model and simplifies the calculation of texture recognition. Also, we propose a novel simulation experiment which can improve the result of image retrieval.