Semantic image retrieval system based on object relationships

Semantic-based image retrieval has recently become popular as an avenue to improve retrieval accuracy. The “semantic gap” between the visual features and the high-level semantic features could be narrowed down by utilizing this kind of retrieval method. However, most of the current methods of semantic-based image retrieval utilize visual semantic features and do not consider spatial relationships. We build a system for content-based image retrieval from image collections on the web and tackle the challenges of distinguishing between images that contain similar objects, in order to capture the semantic meaning of a search query. In order to do so, we utilize a combination of segmentation into objects as well as the relationships of these objects with each other.