Multiple-Kernel Local-Patch Descriptor

Mukundan, Arun; Tolias, Giorgos; Chum, Ondřej

doi:10.5244/c.31.184

Cited by 12 publications

(8 citation statements)

References 31 publications

(65 reference statements)

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Other hand-crafted descriptors considered for the evaluation are Local Intensity Order Pattern (LIOP) [31], employing intensity order pooling and histograms computed on the relative order of neighbor pixels to achieve rotation invariance, and the Multiple-Kernel Local-Patch Descriptor (MKD) [24], using alternative kernels for defining histograms. For both descriptors, optimized data-driven versions exist, exploiting among others Principal Component Analysis (PCA) to achieve better matching results while reducing the associated vector dimensions.…”

Section: Recent State-of-the-art Local Descriptorsmentioning

confidence: 99%

An Evaluation of Recent Local Image Descriptors for Real-World Applications of Image Matching

Bellavia

Colombo

2019

2019 16th International Conference on Machine Vision Applications (MVA)

View full text Add to dashboard Cite

This paper discusses and compares the best and most recent local descriptors, evaluating them on increasingly complex image matching tasks, encompassing planar and non-planar scenarios under severe viewpoint changes. This evaluation, aimed at assessing descriptor suitability for real-world applications, leverages the concept of Approximated Overlap error as a means to naturally extend to non-planar scenes the standard metric used for planar scenes. According to the evaluation results, most descriptors exhibit a gradual performance degradation in the transition from planar to non-planar scenes. The best descriptors are those capable of capturing well not only the local image context, but also the global scene structure. Data-driven approaches are shown to have reached the matching robustness and accuracy of the best hand-crafted descriptors.

show abstract

Section: Recent State-of-the-art Local Descriptorsmentioning

confidence: 99%

An Evaluation of Recent Local Image Descriptors for Real-World Applications of Image Matching

Bellavia

Colombo

2019

2019 16th International Conference on Machine Vision Applications (MVA)

View full text Add to dashboard Cite

show abstract

“…Results are presented for SIFT, sGLOH2 and their variants described in the previous sections (the "P" and "R" prefixes indicate respectively q N,3 (y i ) and R(y i ), RSIFT being a synonymous of RootSIFT), and for a large representative of recent state-of-the-art descriptors. These include: (a) Hand-crafted descriptors: MKD [50], LIOP [65], BRISK [35], FREAK [2], (b) Non-deep data-driven descriptors: MIOP [65], RFD [20], BRIEF [16], LATCH [36], BinBoost [60], ORB [51], BGM [61], LDAHash [55], (c) Deep descriptors: DeepDesc [53], HardNet [45], L2-Net and its binary variant BiL2-Net [57], Geo-Desc [41] and DOAP [27], together with its binary variant BiDOAP. Several versions of the above descriptors are tested, among those proposed by their authors.…”

Section: Resultsmentioning

confidence: 99%

“…The property upon which the popular Scale Invariant Feature Transform (SIFT) descriptor [40] and its inexhaustible crowd of descendants [3,9,19,22,26,31,48] are based is gradient orientation. Other histogrambased descriptors use pixel ordering [65], Haar wavelets [8], convolutions with Gaussian [58] or other kernels [50], and binary pixel comparisons [28]. Data-driven descriptors are those whose structure and design have been tuned and refined according to training data.…”

Section: Introductionmentioning

confidence: 99%

Is There Anything New to Say About SIFT Matching?

Bellavia

Colombo

2020

Int J Comput Vis

View full text Add to dashboard Cite

SIFT is a classical hand-crafted, histogrambased descriptor that has deeply affected research on image matching for more than a decade. In this paper, a critical review of the aspects that affect SIFT matching performance is carried out, and novel descriptor design strategies are introduced and individually evaluated. These encompass quantization, binarization and hierarchical cascade filtering as means to reduce data storage and increase matching efficiency, with no significant loss of accuracy. An original contextual matching strategy based on a symmetrical variant of the usual nearest-neighbor ratio is discussed as well, that can increase the discriminative power of any descriptor. The paper then undertakes a comprehensive experimental evaluation of state-of-the-art hand-crafted and datadriven descriptors, also including the most recent deep descriptors. Comparisons are carried out according to several performance parameters, among which accuracy and space-time efficiency. Results are provided for both planar and non-planar scenes, the latter being evaluated with a new benchmark based on the concept of approximated patch overlap. Experimental evidence shows that, despite their age, SIFT and other handcrafted descriptors, once enhanced through the proposed strategies, are ready to meet the future image matching challenges. We also believe that the lessons learned from this work will inspire the design of better hand-crafted and data-driven descriptors.

show abstract

“…This work is an extension of our earlier conference publication [41]. In addition to the earlier version, we propose unsupervised whitening with shrinkage, give extra insight about its effect on patch similarity, present extended comparisons of different whitening variants and provide a proof justifying the absence of regularized concatenation.…”

Section: Introductionmentioning

confidence: 91%

Understanding and Improving Kernel Local Descriptors

et al. 2018

Self Cite

View full text Add to dashboard Cite

We propose a multiple-kernel local-patch descriptor based on efficient match kernels from pixel gradients. It combines two parametrizations of gradient position and direction, each parametrization provides robustness to a different type of patch mis-registration: polar parametrization for noise in the patch dominant orientation detection, Cartesian for imprecise location of the feature point. Combined with whitening of the descriptor space, that is learned with or without supervision, the performance is significantly improved. We analyze the effect of the whitening on patch similarity and demonstrate its semantic meaning. Our unsupervised variant is the best performing descriptor constructed without the need of labeled data. Despite the simplicity of the proposed descriptor, it competes well with deep learning approaches on a number of different tasks.

show abstract

Multiple-Kernel Local-Patch Descriptor

Cited by 12 publications

References 31 publications

An Evaluation of Recent Local Image Descriptors for Real-World Applications of Image Matching

An Evaluation of Recent Local Image Descriptors for Real-World Applications of Image Matching

Is There Anything New to Say About SIFT Matching?

Understanding and Improving Kernel Local Descriptors

Contact Info

Product

Resources

About