A novel content-based image retrieval approach based on attention-driven model

Visual attention plays a vital role for humans to understand a scene by intuitively emphasizing some focused objects, and recent work in the model of visual attention has demonstrated that a purely bottom-up approach to identify salient regions within an image can be successfully applied to diverse problems. Being aware of this, a novel approach of extracting objects of interest (OOIs) based on attention-driven in an image is proposed. In this approach, the modified Itti-Koch model (M-Itti-Koch) of visual attention is used to find salient peaks, and then if these peaks overlap with regions generated by EM (expectation-maximization) algorithm, we proceed to extract attentive object around that point. Only these objects are considered for the next step, feature extraction and match. This attention-driven model used for CBIR provides a promising performance, as compared with some current peer systems in the literature.