Citation Context

...e postponed until x6. The method is based on evaluating well-normalized local histograms of image gradient orientations in a dense grid. Similar features have seen increasing use over the past decade =-=[4,5,12,15]-=-. The basic idea is that local object appearance and shape can often be characterized rather well by the distribution of local intensity gradients or 1 colour Normalize gamma & gradients Compute into ...

Citation Context

...e the object detection framework of Zhang et al. [18]. Given an image, we use two complementary scale-invariant local region detectors to extract salient image structures: the Harris-Laplace detector =-=[12]-=- responds to corner-like regions and the Laplacian detector [10] extracts blob-like regions. To compute appearancebased features of the patches extracted by the detectors, we use a combination of the ...

Citation Context

...s) with extended Gaussian kernels for classication. 2.1. Detection of interest points We use two complementary local region detectors to extract salient image structures: the Harris-Laplace detector =-=[17]-=- responding to corner-like regions and the Laplacian detector [14] extracting blob-like regions. These two detectors are invariant to scale transformations, i.e., they output circular regions at a cer...

Citation Context

...be in subsection 2.3, how to construct a classier that evaluates the condence of a shape mask to lie on an object. 2.1. Sparse local features and alignment Given an image, we use the Harris-Laplace =-=[13]-=- or the Laplacian interest point detector [9] to nd a sparse set of salient image features. In our experiments the HarrisLaplace detector usually produces comparable results with a lower number of de...

Citation Context

...tial matching cost. 2. Components of the representation 2.1. Image representation We use two complementary local region detector types to extract salient image structures: The Harris-Laplace detector =-=[15]-=- responds to corner-like regions, while the Laplacian detector [12] extracts blob-like regions. These two detectors are invariant to scale transformations only they output circular regions at a cert...

Citation Context

.... A. Related work Performance evaluation has gained more and more importance in computer vision [7]. In the context of matching and recognition several authors have evaluated interest point detectors =-=[14, 30, 33, 39]-=-. The performance is measured by the repeatability rate, that is the percentage of points simultaneously present in two images. The higher the repeatability rate between two images, the more points ca...

Citation Context

...itive should be both discriminative between different types of objects, and repeatable within a specic class of object. Previous research has used single features such as SIFT [12], corner detectors =-=[9, 14]-=-, maximally stable extremal regions [13, 15], and salient regions [10]. The use of single features provides a high-level of repeatability. However, the occurrence of these features is not unique to a ...

Citation Context

...ved, and executables are available on our web page http://www.inrialpes.fr/ lear/downloads.html. The different detectors and descriptors were compared in our evaluation of interest points and regions =-=[7]-=-. The images used in this evaluation are also available on the web at http://www.inrialpes.fr/movi/people/Mikolajczyk/Database. 6 New results 6.1 Image description Contributed by: Cordelia Schmid, Mic...

Citation Context

...imagery with varying viewpoints and illumination conditions. III. INFORMATIVE LOCAL DESCRIPTORS Research on visual object detection has recently focused on the development of local interest operators =-=[7]-=-, [8], [9], [1] and the integration of local information into robust object recognition [10], [9], [1]. Recognition from local information serves several purposes, such as, improved tolerance to occlu...

Citation Context

... Their main advantage is to more efciently detect local symmetries against background clutters in images that are not globally symmetric. Moreover, the recent development of local invariant features =-=[11, 14, 15]-=- has brought about signicant progress in this approach. Tuytelaars et al. [19] presented a method for the detection of regular repetitions of planar patterns under perspective skew using a geometric ...