Citation Context

...al relationships (such as epipolar geometry). Proceedings of the Ninth IEEE International Conference on Computer Vision (ICCV 2003) 2-Volume Set 0-7695-1950-4/03 $17.00 © 2003 IEEE 1 Examples include =-=[5, 6, 8, 11, 13, 12, 14, 16, 17]-=-. We explore whether this type of approach to recognition can be recast as text retrieval. In essence this requires a visual analogy of a word, and here we provide this by vector quantizing the descri...

Citation Context

...ply scale selection [8] to select characteristic points. These points are invariant to scale changes and allow matching and recognition in the presence of large scale factors. Tuytelaars and Van Gool =-=[16]-=- detect affine invariant regions based on image intensities. However, the number of such regions in an image is limited and depends on the content. They use colour descriptors computed for these regio...

Citation Context

...hmid and by Schaffalitzky and Zisserman; a detector of ‘maximally stable extremal regions’, proposed by Matas et al. [21]; an edge-based region detector [45] and a detector based on intensity extrema =-=[47]-=-, proposed by Tuytelaars and Van Gool; and a detector of ‘salient regions’, proposed by Kadir, Zisserman and Brady [12]. The performance is measured against changes in viewpoint, scale, illumination, ...

Citation Context

...r first contribution. Our method does away with having to compute ad hoc descriptors, rectify patches, and search for nearest-neighbors, which are time-consuming steps common in earlier methods [28], =-=[29]-=-, [3], [19], [17], [27], [15], [21]. Another contribution is to show that, in this context, a simple and fast keypoint detector suffices for operation even under large perspective and scale variations...

Citation Context

...camera registration to object detection. Since the pioneering work by Schmid and Mohr [1], the algorithms have become ever more robust to scale, viewpoint, illumination changes and partial occlusions =-=[2, 3, 4, 5, 6, 7]-=-. These wide baseline matching methods, however, are typically designed to match two images but do not take advantage of the fact that, for object detection and pose estimation purposes, both a 3D obj...

Citation Context

...rs. By using di#erent numbers of points we can form feature descriptors which are invariant to any 2D projective transformation (similarity, a#nity or homography). Similar ideas have been proposed by =-=[12]-=-, but they have su#ered from the lack of scale-invariant interest point detectors in their implementation. Groups of interest points which are nearest neighbours in scale-space are used to calibrate t...

Citation Context

...variant point detector based on the Harris detector, where the affine transformation that makes equal the two eigen values of the auto correlation matrix is evaluated to rectify the patch appearance. =-=[134]-=- achieves such invariance by fitting an ellipse to the local texture. [87] proposes a fast algorithm to extract Maximally Stable Extremal Regions demonstrated in a live demo. [92] gives a good summary...

Citation Context

... more than 25-30 ◦ . 1 Introduction Detecting and matching specific visual features across different images has been shown to be useful for a diverse set of visual tasks including stereoscopic vision =-=[1, 2]-=-, vision-based simultaneous localization and mapping for autonomous vehicles [3], mosaicking images [4] and recognizing objects [5, 6]. This operation typically involves three distinct steps. First a ...

Citation Context

...icient computation at every pixel in the image. Local image descriptors have already been used in dense matching, though in a more traditional way, to match only sparse pixels that are feature points =-=[31]-=-, [19]. In [27], [33], these matched points are used as anchors for computing the full reconstruction. Yao and Cham [33] propagate the disparities of the matched feature points to their neighbors, whi...