Conceptual vision keys for consumer product images

In the consumer world, the ever growing image repositories in online shopping, consumer products images, consumer photos and video collections have resulted great demand of a system which can accurately retrieve similar images from image database. For this purpose, we propose a new concept of vision key for retrieving consumer product images. In our system, rather than considering an image as a whole, we consider it as a set of regions or sub-images with completely different semantic meanings. By using the properties of equivalence classes in the Markov chain, we first perform image segmentation and initial pixel grouping process. We then establish vision keys by using a Markov stationary feature. Finally, in the retrieval phase, users can interactively search candidate images which contain vision keys. In order to confirm the efficiency of our proposed method, we present the experimental results achieving on higher accuracy rates.