Presentation is loading. Please wait.

Presentation is loading. Please wait.

A Fast Local Descriptor for Dense Matching Engin Tola, Vincent Lepetit, Pascal Fua Computer Vision Laboratory, EPFL Reporter : Jheng-You Lin 1.

Similar presentations


Presentation on theme: "A Fast Local Descriptor for Dense Matching Engin Tola, Vincent Lepetit, Pascal Fua Computer Vision Laboratory, EPFL Reporter : Jheng-You Lin 1."— Presentation transcript:

1 A Fast Local Descriptor for Dense Matching Engin Tola, Vincent Lepetit, Pascal Fua Computer Vision Laboratory, EPFL Reporter : Jheng-You Lin 1

2 Introduction DAISY Computation Results Conclusion Outline

3 Wide-base line matching propose : SIFT 、 GLOH 、 SURF… (histogram based descriptor) – Good performance and robustness to image transformations. – High computational cost and sensitivity to occlusions. Purpose – Design a descriptor that is as robust as SIFT or GLOH but can be computed much more effectively and handle occlusions. Introduction

4 No velty – introduces DAISY local image descriptor Introduction (cont.)

5 No velty – introduces DAISY local image descriptor Introduction (cont.) SIFT descriptor is a 3–D histogram in which two dimensions correspond to image spatial dimensions and the additional dimension to the image gradient direction (normally discrete into 8 bins)

6 No velty – introduces DAISY local image descriptor Introduction (cont.) * S. Winder and M. Brown. Learning Local Image Descriptors in CVPR’07 Improved performance : + Precise localization + Rotational Robustness

7 No velty – introduces DAISY local image descriptor Introduction (cont.) Replacing weighted sums by convolutions

8 DAISY Computation

9 First compute gradient magnitude layers in different orientations

10 DAISY Computation Then, apply convolution with a Gaussian kernel to pre-compute the histograms for every point

11 DAISY Computation

12

13

14 The computation mostly involves 1D convolutions, which is fast.

15 DAISY Computation Rotating the descriptor only involves reordering the histograms. …

16 DAISY Computation Rotating the descriptor only involves reordering the histograms. …

17 DAISY Computation Computation Time Comparison(in seconds)

18 DAISY Computation The full DAISY descriptor D(u, v) : The descriptor of the same point that is close to an occlusion would be very different. Normalize to unit norm

19 Results Laser scanDAISYSIFT SURFPixel DifferenceNCC

20 Results baseline increase block Error threshold : Top : 10% Middle : 5% Bottom : 1% DAISYSIFTSURF NCC SURFPixel Difference

21 Results Using low-resolution of the Brussels images[24] 768x510 (2048x1360 origin) [24] Combined Depth and Outlier Estimation in Multi-View Stereo, CVPR’06

22 Results Using low-resolution of the Rathaus images[25] 768x512 (3072x2048 origin) The holes are caused by the fact that a lot of the texture is not visible. [25] Dense Matching of Multiple Wide-Baseline Views, ICCV’03

23 Results Input imagesVirtual viewSynthesized

24 Results Virtual viewSynthesizedDAISYNCC

25 Efficient descriptor and produces good reconstructions. Can handle low quality imagery Conclusion


Download ppt "A Fast Local Descriptor for Dense Matching Engin Tola, Vincent Lepetit, Pascal Fua Computer Vision Laboratory, EPFL Reporter : Jheng-You Lin 1."

Similar presentations


Ads by Google