Region and texture based effective image extraction

Images can be well defined by the primitive shape and texture features. In this paper we combined these features in a novel way. We presented a scheme of the local interest point detection and their global description in a more meaningful way. We extracted the image signatures by assembling the interest points at different levels of the representation. For this, at first shape feature are collected by grouping the connected pixels to create the regions based on binary intensity threshold. Histogram of oriented gradients is used to describe the features for the detected interest points returned for maximally stable regions. These signatures are combined with rotation invariant texture features extracted by using uniform local binary pattern after applying the proposed reordering algorithm. The algorithm intakes the number of observations for the region and texture features and returns the compact dimensions as input for the limited principal components computation. Our proposed technique is experimented against the existing research methods on Corel-100, Caltech-101 and Caltech-256 datasets and outperforms in many image categories. Experimentation results show that a combination of local and global features strengthens the capability of the proposed method to retrieve the foreground and background objects. Furthermore, feature description using sliding window fashion makes this approach more robust to object recognition.