Duplicate image retrieval with cascade similarity measurements

作者：Cao Yudong^*

来源：International Journal of Advancements in Computing Technology, 2012, 4(10): 228-237.

DOI：10.4156/ijact.vol4.issue 10.27

• Data sets
• Detection filter
• Detection methods
• Duplicate detection
• Duplicate retrieval
• Epipolar geometry
• False negatives
• False positive
• Local feature
• Maximally Stable Extremal Regions
• Scale invariant feature transforms
• Second level
• Similarity measure
• Similarity measurements
• Spatial location
• Storage spaces

In this paper, a cascade detection method using two different similarity measurements is presented. Given a query, the first level detection filters out the most of non-duplicate images and remains most of duplicate images. Then, the second level detection with epipolar geometry is applied only on the remained images so it can remove large number of false positives and false negatives further. In the second level detection, the spatial location relationship is considered in the similarity measure. In this experiment we represent images with a combination of local features, i.e., SIFT (Scale-invariant feature transform) and MSER (Maximally Stable Extremal Region) which increases discrimination and saves storage space than using SIFT alone. Finally the experiments show the promising results of the proposed method on the benchmarked Flickr dataset and INRIA dataset.

• 出版日期2012
• 单位辽宁大学