Using fuzzy rough feature selection for image retrieval system

Feature selection is an important step in processing the images especially for applications such as content based image retrieval. In large multimedia databases, it may not be practical to search through the entire database in order to retrieve similar images from a query. Good data structures for similarity search and indexing are needed, and the existing data structures do not scale well for the high dimensional multimedia descriptors. Thus feature selection is an important step. Fuzzy rough feature selection method has many advantages in determining the relevant features. In this paper, five feature selection methods are compared with the fuzzy rough method. These five feature selection methods are Relief-F, Information Gain, Gain Ratio, OneR and the statistical measure χ². The main purpose of the comparison is to rank the image features and see which method provides better results. An image retrieval dataset (COREL dataset) was used in the comparison. In order to evaluate the performance of the six methods, ranking of the important features is defined. This is then used to compare with the automated ranking produced by the aforesaid feature selection methods. Results show that the retrieval system using fuzzy rough feature selection has better retrieval accuracy and provide good Precision Recall performance. The advantages of the use of fuzzy rough feature selection will also be discussed in the paper.