Image Retrieval Based on Fuzzy Set Theory and Development of an Experimental System.

近年来，随着多媒体数据库和超文本研究的发展，图像检索技术迅速成为这些领域的重要技术。在过去的图像检索中，已经使用了一些方法。例如，使用给定的图片进行查询，检索结果是图像数据库中最相似的图片的相似图像检索方法，以及使用网络索引的方法。在图像索引和检索中，模式识别技术和认知科学的产物是基本的和有用的技术。另一方面，图像数据本身具有一定的不确定性或模糊性，将模糊集理论应用于图像检索是很自然的，但目前还没有提出模糊图像检索的方法。本文提出了一种基于模糊集理论的图像检索新方法，并实现了一个实验图像检索系统。研究总结如下：1。针对人类的图像记忆过程，提出了命题网络模型。在此模型的基础上，利用模糊命题集作为图像指标。我们假设查询由一组具有相同图像数据库索引结构的模糊命题组成。使用这种类型的查询，我们可以制作各种用户界面，其中用户查询可以是一组关键字或自然语言表达式或参考图像。提出了查询与索引匹配度的模糊度量模型，并给出了一种有效的匹配度计算算法。该度作为图像数据库检索过程中的模糊隶属度值。我们利用这种方法实现了一个实验性的图像检索系统。

Recently, as the result of the development of the researches on multimedia databases and hypertexts, image retrieval is rapidly becoming more important technique in those fields. In the past, some methods have been used for the image retrieval. For example, the similar image retrieval method, in which a given picture is used for a query and the retrieved result is the most similar picture in a image database, and the method of using network index are well known. On image indexing and retrieving, the pattern recognition techniques and the products of cognitive science are basic and useful techniques generally. On the other hand, image data has some sort of uncertainty or fuzziness in itself and it is natural to apply fuzzy set theory to image retrieval, but the method of fuzzy image retrieval is not proposed yet.In this research, we proposed new method of image retrieval based on fuzzy set theory and implement an experimental image retrieval system. The summary of the research is shown below :1. The propositional network model was proposed for the image memory process of human beings. Based on this model, the set of fuzzy propositions are used as the image index.2. We assumed that the query is composed a set of fuzzy propositions that has the same structure of image database indexes. Using this type of query, we can make various user interfaces in which user query can be a set of keywords or an expression in natural language or a reference image.3. We proposed a fuzzy model for measuring a matching degree between a query and a index and produce an algorithm which calculates a matching degree efficiently. This degree is used as a fuzzy membership value on the retrieving process of the image database.4. We implemented an experimental image retrieval system using this method.