#### DMCA

## Fast approximated SIFT (2006)

### Cached

### Download Links

- [www.icg.tugraz.at]
- [www.icg.tugraz.at]
- [nichol.as]
- [nichol.as]
- [www.icg.tugraz.at]
- [www.icg.tugraz.at]
- [www.vision.ee.ethz.ch]
- DBLP

### Other Repositories/Bibliography

Venue: | IN 7TH ASIAN CONFERENCE OF COMPUTER VISION |

Citations: | 34 - 5 self |

### Citations

8751 | Distinctive image features from scale-invariant keypoints. IJCV,
- Lowe
- 2004
(Show Context)
Citation Context ... segmentation of the image and due to the local nature they are robust to occlusions. Local approaches have demonstrated considerable success in a variety of applications, like recognition of objects =-=[1]-=-, wide-base line stereo [4], robot navigation [5], image retrieval [6, 7], building of panoramas [8], etc. Probably the most popular and widely used local approach is the DoG detector with the SIFT de... |

3205 | Rapid object detection using a boosted cascade of simple features.
- Viola, Jones
- 2001
(Show Context)
Citation Context ...n (DoM ) images insteadsof Difference-of-Gaussians (DoG). This DoM images can be computed very efficiently by using a box filter in combination with an integral image as introduced by Viola and Jones =-=[11]-=- (capturing the main idea of [12]). Once the integral image is computed, it allows to compute the mean within a rectangular region in constant time independent of the size of the region. This property... |

2662 | Object Recognition from Local Scale-Invariant Features[C].International Conferenceon Computer Vision, Corfu,
- LOWE
- 1999
(Show Context)
Citation Context ...epend on the size of the input image W × H. However, as can be seen in Figure 2(a), the computational costs are higher than for the separable Gaussian for a kernel size of 7×7 (as proposed by Lowe in =-=[14]-=-). A similar result holds for recursive Gaussian filters which allow convolution in constant time but are still computationally more demanding for small filter kernels. Table 2. Comparison of various ... |

1742 | Aperformance evaluation of local descriptors
- Mikolajczyk, Schmid
- 2005
(Show Context)
Citation Context ...ransport, Innovation and Technology under P-Nr. I2-2-26p VITUS2 and by the Austrian Joint Research Project Cognitive Vision under projects S9103-N04 and S9104-N04, the EC funded NOE MUSCLE IST 507572s=-=[9]-=-) demonstrate the excellent performance of the method compared to other approaches. The DoG detector detects blobs in the Laplacian scale space. The SIFT descriptor is basically a histogram (in fact 1... |

1451 | Scale & affine invariant interest point detectors
- Mikolajczyk, Schmid
- 2004
(Show Context)
Citation Context ...ns. This is especially suited for our approach because we always compute the descriptors on the original resolution. Consequently, we take advantage of using the whole information of the input image. =-=(3)-=-s(a) Different filtering techniques for a 7 × 7 filter kernel. (b) Conventional and integral technique for orientation histogram computation. Fig. 2. Comparison of computational costs for detector (le... |

996 | Robust wide baseline stereo from maximally stable extremal region
- Matas, Chum, et al.
- 2002
(Show Context)
Citation Context ...and due to the local nature they are robust to occlusions. Local approaches have demonstrated considerable success in a variety of applications, like recognition of objects [1], wide-base line stereo =-=[4]-=-, robot navigation [5], image retrieval [6, 7], building of panoramas [8], etc. Probably the most popular and widely used local approach is the DoG detector with the SIFT descriptor as proposed by Low... |

571 | Pca-sift: A more distinctive representation for local image descriptors
- Ke, Sukthankar
- 2004
(Show Context)
Citation Context ... be computed fast. Due to the high popularity of SIFT, it is no surprise that several variants and extensions of SIFT have been proposed. For example Ke and Sukthankar proposed the so called PCA-SIFT =-=[10]-=- that applies Principal Components Analysis (PCA) to the normalized gradient patch. The Gradient location and orientation histogram (GLOH) [9] changes SIFTs location grid and uses PCA to reduce the si... |

400 | Indexing based on scale invariant interest points
- Mikolajczyk, Schmid
- 2001
(Show Context)
Citation Context ...o the histogram which leads to the complexity O(N 2 ) for a squared region wheresN corresponds to the window size. In addition the computational costs for a squared region is k · N 2 · (cadd + cmult) =-=(2)-=- where k corresponds to the number of histograms, cadd represent costs for an addition and cmult are the costs for a multiplication. Considering the integral histogram computation illustrated in Algor... |

290 | Recognising panoramas
- Brown, Lowe
- 2003
(Show Context)
Citation Context ...es have demonstrated considerable success in a variety of applications, like recognition of objects [1], wide-base line stereo [4], robot navigation [5], image retrieval [6, 7], building of panoramas =-=[8]-=-, etc. Probably the most popular and widely used local approach is the DoG detector with the SIFT descriptor as proposed by Lowe [1]. SIFT has been used with success in all of the above mentioned appl... |

216 | Integral histograms: a fast way to extract histograms in cartesian spaces. In: CVPR
- Porikli
- 2005
(Show Context)
Citation Context ... with 8 bins is computed and concatenated to form a single feature vector. Since orientation histograms form the basic computation for the descriptor this leads to the idea to use integral histograms =-=[15]-=-. Integral histograms are an extension of integral images using for each histogram bin (e.g. orientation) a separate integral image. Once the integral orientation histogram is computed, histograms can... |

109 | Content-based image retrieval based on local affinely invariant regions
- Tuytelaars, Gool
- 1999
(Show Context)
Citation Context ... to occlusions. Local approaches have demonstrated considerable success in a variety of applications, like recognition of objects [1], wide-base line stereo [4], robot navigation [5], image retrieval =-=[6, 7]-=-, building of panoramas [8], etc. Probably the most popular and widely used local approach is the DoG detector with the SIFT descriptor as proposed by Lowe [1]. SIFT has been used with success in all ... |

68 |
An efficient parts-based near-duplicate and sub-image retrieval system
- Ke, Sukthankar, et al.
- 2004
(Show Context)
Citation Context ... to occlusions. Local approaches have demonstrated considerable success in a variety of applications, like recognition of objects [1], wide-base line stereo [4], robot navigation [5], image retrieval =-=[6, 7]-=-, building of panoramas [8], etc. Probably the most popular and widely used local approach is the DoG detector with the SIFT descriptor as proposed by Lowe [1]. SIFT has been used with success in all ... |

55 |
de Weijer, “Fast anisotropic gauss filtering
- Geusebroek, Smeulders, et al.
- 2003
(Show Context)
Citation Context ...sponse can be computed, independent of its size, with 4 memory accesses, 3 additions and a single multiplication which is needed for normalizing the box region. In Table 2 which has been adapted from =-=[13]-=-, we Algorithm 1 Integral image computation // pre-computation for each image point do Propagate integral image {1 addition} Increase value {1 addition} end for // apply box filter with a given kernel... |

52 | Local and global localization for mobile robots using visual landmarks
- Se, Lowe, et al.
- 2001
(Show Context)
Citation Context ...ature they are robust to occlusions. Local approaches have demonstrated considerable success in a variety of applications, like recognition of objects [1], wide-base line stereo [4], robot navigation =-=[5]-=-, image retrieval [6, 7], building of panoramas [8], etc. Probably the most popular and widely used local approach is the DoG detector with the SIFT descriptor as proposed by Lowe [1]. SIFT has been u... |

37 | Boxlets: a fast convolution algorithm for signal processing and neural networks
- Simard, Bottou, et al.
- 1999
(Show Context)
Citation Context ...ence-of-Gaussians (DoG). This DoM images can be computed very efficiently by using a box filter in combination with an integral image as introduced by Viola and Jones [11] (capturing the main idea of =-=[12]-=-). Once the integral image is computed, it allows to compute the mean within a rectangular region in constant time independent of the size of the region. This property allows fast box filtering and ca... |

31 | Object class recognition using multiple layer boosting with heterogenous features - Zhang, Yu, et al. - 2005 |