Use of multiple tools for texture based image retrieval is proposed. Each tool may be composed of a classifier, a feature vector reduction part and a distance metric. The idea behind the proposed scheme is that computed features do not correspond exactly to the visual attributes and thus, computational result of similarity between two images is not unique. One solution is to run parallelly many tools which will select a subset of features from a pool and which will use a distance metric among many alternatives. The results of the tools can then be combined by a majority voting. The proposed system is experimented on a sample image database. The results show that the combination of search by different approaches improves the system performance. Further improvements will be incorporated to the system by additional attributes such as color.
展开▼
