Inria IMEDIA2's Participation at ImageCLEF 2012 Plant Identification Task
Abstract
This paper describes the participation of Inria IMEDIA2 Team, within the Pl@ntNet project4, at the ImageCLEF2012 plant identi cation task. The runs used very distinct approaches, sometimes relying on similar extracted features. For Scan and Scan-like categories, the rst two runs combine distinct local and contour approaches in two ways (late and early fusion), while the third run explores the learning capacity of a multi-class SVM technique on a contour based descriptor. For Photo- graph our runs used local features positioned towards the center of the image to reduce the impact of background features. In the second run, an automatic segmentation with a rejection criterion was attempted. In the third run, points were associated with interesting zones. In general, even if they were distinct, the methods used performed very well.
Origin : Publisher files allowed on an open archive
Loading...