Active contours: Application to plant recognition

The problem we address in this paper is object segmentation applied to plant recognition. The image can contain one or more plants on a natural background. More precisely, we aim to segment flowers. This approach poses several challenges, such as texture, multiple colors that form one object, natural background, non-homogeneous regions, etc. We propose an approach that adapts the Chan-Vese model, [1], in order to use it with a fast level-set approximation algorithm, [2]. Our approach presents ongoing work towards flower segmentation and recognition. In this paper we present the main outcome of our research on segmentation problem (curve evolution from Chan-Vese model with a recent and faster approach, [2], and its optimization at implementation level) and on recognition problem (identifying flower species). At recognition level, we studied five features that may be extracted during segmentation. Experiments have been carried out over the Oxford Flowers dataset, [3].