CLEF2008 Image Annotation Task: an SVM Confidence-Based Approach
This paper presents the algorithms and results of our participation to the medi- cal image annotation task of ImageCLEFmed 2008. Our previous experience in the same task in 2007 suggests that combining multiple cues with diﬀerent SVM-based approaches is very eﬀective in this domain. Moreover it points out that local features are the most discriminative cues for the problem at hand. On these basis we decided to integrate two diﬀerent local structural and textural descriptors. Cues are combined through simple concatenation of the feature vectors and through the Multi-Cue Ker- nel. The trickiest part of the challenge this year was annotating images coming mainly from classes with only few examples in the training set. We tackled the problem on two fronts: (1) we introduced a further integration strategy using SVM as an opinion maker. It consists in combining the ﬁrst two opinions on the basis of a technique to evaluate the conﬁdence of the classiﬁer’s decisions. This approach produces class labels with “don’t know” wildcards opportunely placed; (2) we enriched the poorly populated training classes adding virtual examples generated slightly modifying the original images. We submitted several runs considering diﬀerent combination of the proposed techniques. Our team was called “idiap”. The run using jointly the low cue- integration technique, the conﬁdence-based opinion fusion and the virtual examples, scored 74.92 ranking ﬁrst among all submissions.
CLEF 2008 Working Notes
Record created on 2010-02-11, modified on 2016-08-08