Pre-processing image database for efficient Content Based Image Retrieval

Content Based Image Retrieval (CBIR) has been a challenging area of research for more than a decade. In this area of research, selection of features to represent an image in the database is still an unresolved issue. Unfortunately, the existing solutions regarding the problem are only focusing on the relevance feedback techniques to improve the count of similar images related to a query from the raw image database. These approaches are inefficient and inaccurate to query the image. We propose a new efficient technique to solve these problems by exploiting a new strategy called preprocessing image database using k-means clustering and genetic algorithm. This technique utilizes several features of the image, such as color, edge density, boolean edge density and histogram information as the input of retrieval. Furthermore, several performance metrics, such as confusion matrix, precision graph and F-measures, have also been used in measuring the accuracy of the proposed technique. The experiment results show that the clustering purity in more than half of the clusters has been above 90 percent purity.