Content-Based Image (object) Retrieval with Rotational Invariant Bag-of-Visual Words representation

While Bag of Visual Word (BoW) approaches have achieved better retrieval performance and known as a precise representation, the performance is limited as it ignores spatial information. This paper presents a simple yet effective Rotation Invariant Bag of Visual Words (RIBoW) approach which encodes spatial information to achieve effective Content-Based Image Retrieval (CBIR) especially in object based retrieval. RIBoW approach uses circular image decomposition in combination with a simple shifting operation to achieve invariance in rotation using global image descriptors. Feature vectors are histograms of visual words in the traditional BoW approach. The given feature indices and cluster indices are together used to generate the signatures of the proposed method. The retrieval quality of the proposed approach is empirically evaluated using two benchmark datasets for image classification (a subset of the Corel dataset) and compared. Proposed approach is further validated with Caltech 256 dataset. The performance of the proposed approach confirms its effectiveness and robustness for rotation invariant image retrieval by transcending retrieval performance when comparing with BoW approaches.