In this work, SIFT features are revisited for their use in two applications of computer vision for plantanalysis. The first application is the reconstruction of 3D models of plants through tracking homologuepoints in successive intensity images. The second application is to provide a new global descriptor thatgives a measure of the level of self-similariy of foliage for plants of different architectures and foliarappearance. In order to properly exploit SIFT descriptors in relation to these applications, we discuss twoaspects of the classical SIFT keypoint matching practice. On the one hand we propose to match detectedkeypoints based on a scale criterion. On the other hand, we drop the ratio rule while matching keypointsin two images and propose the use of a spatial proximity filter instead.