Image-Based Accurate Object Retrieval Combined with Color Invariant

Finding information based on an object´s profile is very useful when exact keywords for the object are unknown. Current image retrieval system all ignores the color information, for example we want to find a super-star with a piece of red petticoat, or we want to a red flower with white background. They all cannot give the desired results in state of the art image search engine. We have developed an object retrieval system that takes images of objects as queries and finds relevant image scenes that contain the objects, which combined with the color information. We supply a query object by selecting a region of the query image, and the system returns a ranked list of images that contain the same object. We use Caltech-101 as test queries. Creating an image-feature vocabulary is a time-consuming process, and it affects the performance. To address those problems we compare many scalable methods for building the vocabulary tree and introduce a fast adaptive method combined with color information, which we show outperforms the current famous systems.