Abstract: The main aim of the proposed system is use image click-through data, which can be viewed as the implicit feedback from users to overcome the intention gap, and further improve the image search performance. This paper presents a novel re-ranking approach, named spectral clustering re-ranking with click-based similarity and typicality using graph base visual saliencing technique (GBVST). The saliencing technique can be used to differentiate foreground and background region according to saliency distribution. To achieve an appropriate similarity dimension, we propose click-based multi-feature similarity learning algorithm. Then based on the learnt click-based image similarity measure, we organized spectral clustering to group visually and semantically similar images into same clusters. The final re-rank list by calculating click-based clusters typicality and within- clusters click-based image typicality in descending order. Our experiment improves the initial image search result.
Keywords: Image search, search re-ranking, saliencing technique, click base similarity, typicality.