Multimedia Search with Pseudo-relevance Feedback

An Adaptive Query Prototype Modeling Method for Image Search Reranking

“…The images reranking methods can be classified into that of classification based [12,25], clustering based [2,3,4,7,14], graph based [8,16], and learning to rerank [22,26,27].…”

“…The negative samples are selected from the images in the bottom of the rank list or from other queries's results. Then a classifier is trained using these positive and negative samples [25]. And the images in the initial list are classified and reranked using the trained classifier.…”

“…What's more, we do the reranking experiments using PRF (Pseudo-relevance Feedback) method [25] on the dataset of WebQueries. We use the color features such as CAC (color auto-correlogram), CCV (color coherence vector), CLD (color layout descriptor), CSD (color structure Table 5.…”

“…Thus, for image reranking, pseudo relevance feedback (PRF) assumption is widely use. For one query, the assumption regards the top-N returned images as positive samples to learn the model of the query [25][26] [12]. And in the model, the top images are usually considered weighted equality without considering the position influence.…”

Boosting Web Image Search by Co-Ranking

“…For example, Yan et al [9] train SVMs whose positive training data are from the query examples, while negative training data are from negative pseudo relevance feedback; however, in the scenario of query by keyword, positive training data is hard to obtain. And Lin et al [10] propose a relevance model to calculate the relevance of each image, which evaluates the relevance of the HTML document linking to the image; however, this model depends on the documents returned by a text web search engine, which may be totally irrelevant with the retrieved images.…”