Fusion at Features Level in CBIR System Using Genetic Algorithm

摘要

The research in content based Image retrieval (CBIR) systems is becoming matured as more and more applications are building up over it. In order to imitate the way human being treat the image management, the conventional way of text based retrieval systems are being replaced by the visual content based systems. The image content has several dominant characteristics like color, texture and it is interesting to see the classification of images on content-basis can be achieved with these features. The color has different features like average, variance and texture can be represented by cooccurrence matrix. This different descriptor for the images can form a combined feature vector. However, in order to have optimum performance and to reduce the feature dimensionality for making system real-time, genetic algorithm (GA) based feature selection is used in this paper. The genetic algorithm can be used at the level of the feature elements selection; where important features are preserved while ignoring remaining We used the database of 10 classes and 100 images in each class for validation of CBIR system. We performed the experiments with and without GA and observed the usefulness of feature optimization. The result shows the effectiveness of these features for content based classification or image retrieval applications.