A fast local descriptor for dense matching

By Engin Tola, Vincent Lepetit, Pascal Fua. 2008 IEEE Conference on Computer Vision and Pattern Recognition, 2008

560 citations recorded by the source corpus.

We introduce a novel local image descriptor designed for dense wide-baseline matching purposes. We feed our descriptors to a graph-cuts based dense depth map estimation algorithm and this yields better wide-baseline performance than the commonly used correlation windows for which the size is hard to tune. As a result, unlike competing techniques that require many high-resolution images to produce good reconstructions, our descriptor can compute them from pairs of low-quality images such as the ones captured by video streams. Our descriptor is inspired from earlier ones such as SIFT and GLOH but can be computed much faster for our purposes. Unlike SURF which can also be computed efficiently at every pixel, it does not introduce artifacts that degrade the matching performance. Our approach was tested with ground truth laser scanned depth maps as well as on a wide variety of image pairs of different resolutions and we show that good reconstructions are achieved even with only two low quality images.

Subfields: Aerospace Engineering, Computer Vision and Pattern Recognition

View Paper: https://www.semanticscholar.org/paper/5ebb1f93f9e4d76ae513ad781c3764e3e62fba40

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Aerospace Engineering

In Computer Vision and Pattern Recognition

Citations in the corpus

SIFT Meets CNN: A Decade Survey of Instance Retrieval (2016, methodology): “ the Hellinger kernel. RootSIFT has been shown to yield superior performance to original SIFT and is now used as a routine in the SIFT-based retrieval models. Ge et al. also test the DAISY descriptor [43] in retrieval. In [44], Tolias et al. propose not to align the detected patches by dominant orie”
Learning to compare image patches via convolutional neural networks (2015, methodology): “Results without global optimization also show that the estimated depth maps contain much more ﬁne details than DAISY.”
Generic 3D Representation via Pose Estimation and Matching (2016, result): “Matching Baselines: We compared our results with the handcrafted features of SIFT [35], Root-SIFT [7], DAISY [55], VIP [61] (which requires the surface normals in the input for which we used the normals from the 3D models), and ASIFT [41].”
Accurate, Dense, and Robust Multiview Stereopsis (2010, background): “See [27] for an example of other photometric discrepancy functions. matching step is to generate a sparse set of patches (possibly containing some false positives).”
LIFT: Learned Invariant Feature Transform (2016, methodology): “Along similar lines, DAISY [3] relies on convolved maps of oriented gradients to approximate the histograms, which yields large computational gains when extracting dense descriptors.”
Accurate, Dense, and Robust Multi-View Stereopsis (2007, background): “3See [27] for an example of other photometric discrepancy functions.”
Pose estimation for category specific multiview object localization (2009, methodology): “The final step confirms the existence of an object within the estimated bounding box, which is done using a single classifier tuned to the estimated viewpoint.”
Scale-invariant heat kernel signatures for non-rigid shape recognition (2010, background): “In image analysis and 2D shape retrieval, typical features are blobs [25] and corners [27], and the default choice for a local descriptor is the scale-invariant feature transform (SIFT) [23] or one of its varieties [2] [41].”
Learning Local Feature Descriptors Using Convex Optimisation (2014, background): “Note that s increases with the distance from the patch centre, which is also specific to certain hand-crafted descriptors, e.g., DAISY [29].”
The power of comparative reasoning (2011, methodology): “Since these methods rely on relative ordering of elements, they are very resilient to noise and variations that do not affect the implicit order.”
Receptive Field Block Net for Accurate and Fast Object Detection (2017, background): “A few shallow descriptors coincidentally make use of such a mechanism to design [36] or learn [1, 39, 31] their pooling schemes, and show good performance in matching image patches.”
Finding meaning on YouTube: Tag recommendation and category discovery (2010, methodology): “Local Features: We employ a custom local descriptor [27, 16] which consists of Gabor responses at different orientations, spatial scales, and offsets with respect to the interest point.”
Face recognition with learning-based descriptor (2010, background): “Recently, face recognition has attracted much research effort [9, 25, 26, 10, 11, 13, 19, 20, 30, 32] due to the progresses of local descriptors [6, 16, 17, 22, 27, 28, 29, 30] and increasing demands of real-world applications, such as face tagging on the desktop [5] or the Internet(1).”
Boundary Detection Using F-Measure-, Filter- and Feature- (F3) Boost (2010, methodology): “As in recent works [13] we compute descriptors densely on the image, thereby capturing the context of any given point, and provide this as an input to a classification algorithm.”