Web Image Search Re-ranking with Click-based Similarity and Typicality

0
436
Web Image Search Re-ranking with Click-based Similarity and Typicality

Web Image Search Re-ranking with Click-based Similarity and Typicality

Abstract of Web Image Search Re-ranking with Click-based Similarity and Typicality

Web Image Search Re-ranking with Click-based Similarity and Typicality.Hundreds of thousands of images are uploaded to the internet with the explosive growth of online social media and the popularity of capture devices, thus, building a satisfying image retrieval system is the key to improve user search experience.

Due to the success of information retrieval, most commercial search engines employ text-based search techniques for image search .

Even though text-based search techniques have achieved great success in document retrieval, text information is often noisy and even unavailable.

In image search re-ranking, besides the well-known semantic gap, intent gap, which is the gap between the representation of users’ query/demand and the real intent of the users, is becoming a major problem restricting the development of image retrieval.

Conclusions

we have studied the issue of leveraging click-through data to reduce the intent gap of image search.

We propose a novel image search re-ranking approach, named spectral clustering re-ranking with click-based similarity and typicality (SCCST).

 With the detection of click-based triplets, we present a novel image similarity measurement, named click-based multi-feature similarity learning (CMSL), which integrates multiple kernel learning into metric learning to learn similarity measure for each feature in a unified space.

Based on the learnt similarity measure, SCCST performs spectral clustering to group visually and semantically similar images into same clusters.