Dynamically Visual Disambiguation of Keyword-based Image Search
This addresses the issue of ambiguous image search results for users relying on web-based learning, though it appears incremental as it builds on existing disambiguation techniques.
The paper tackles the problem of visual polysemy in keyword-based image search by proposing an adaptive multi-model framework for visual disambiguation, which dynamically selects text queries and uses saliency-guided deep multi-instance learning to remove outliers and learn classification models. Experiments show the approach achieves superior performance compared to existing methods.
Due to the high cost of manual annotation, learning directly from the web has attracted broad attention. One issue that limits their performance is the problem of visual polysemy. To address this issue, we present an adaptive multi-model framework that resolves polysemy by visual disambiguation. Compared to existing methods, the primary advantage of our approach lies in that our approach can adapt to the dynamic changes in the search results. Our proposed framework consists of two major steps: we first discover and dynamically select the text queries according to the image search results, then we employ the proposed saliency-guided deep multi-instance learning network to remove outliers and learn classification models for visual disambiguation. Extensive experiments demonstrate the superiority of our proposed approach.