Accurate and contrast invariant junction detection

Gui-Song Xia
CNRS CEREMADE, Univ. Paris-Dauphine, Paris, France

Julie Delon, Yann Gousseau
CNRS LTCI, Telecom ParisTech, Paris, France

Abstract: This paper introduces a generic method for the accurate analysis of junctions, relying on a statistical modeling of normalized image gradients. We analyze junctions as local visual events that do not happen by chance under a background model derived from the a-contrario methodology. The method not only provides thresholds for the detection of junctions, but also enables their accurate characterization, including a precise computation of their type, localization, scale and geometrical configuration. The efficiency of the method is evaluated through various experiments

 

Numerical Results: 

1.   Stability and choice of the threshold

ACJ

Harris

 

Pj on gPb

 

 

Junctions obtained by ACJ (top row), Harris (mid row) and “Pj on gPb” (bottom row). The parameters are fixed for the three images, yielding the same number of detections on the house image (this correspond to epsilon equals 1). Observe that only ACJ prevent from over-detection in textured areas. The color of the junction depends on the NFA value (red corresponds to small values, i.e. very meaningful junctions and blue corresponds to high values).

 

2.   Quantitative evaluation of scale invariance

tested images

Scales of junctions detected by ACJ along  the scale space

The repeatability rate on the scale space

 

Illustration of the scale invariance along the scale space. The first row shows the tested images. The second row shows the scales of junctions detected by ACJ along all the trajectories (the list of detected junctions in the scale space [2] , curves in red) as a function of the zoom factor si  (the abscissa is si). The baselines (y = r si), where r changes from 1 to 90, are displayed in blue. The bottom row presents the repeatability rate of the junctions as a function of the zoom factor.


 

       

Examples of detected junctions along several junction trajectories: each row shows a list of junctions detected at the same relative locations in images, from the coarsest to the finest resolution. In these examples, junctions can be followed along complete trajectories, and their scales remain roughly proportional to the image resolution.

 

3.   Quantitative evaluation of contrast invariance

Repeatability rate of different approaches regarding contrast changes. The curves are averaged over 9 image sequences. Each sequence is obtained by applying different gamma corrections (as specified in the text) to a test image.

 

4.   More detection results

 

  

 

      (a) the characterization of L-, Y- and X-junctions given by the proposed approach;

    (b) result given by Harris-Laplace

Junction characterization on the house image. (a) shows the characteristic scale of L-, Y- and X-junctions given by the proposed approach, and (b) shows the characteristic scale of junctions given by Harris-Laplace. The location of each junctions is indicated by a red cross and circles have a radius equal to the corresponding scale.

 

 

 

  

(a)  the junctions given by the Pj on gPb approach;                                 (b) junctions given by Harris detector          

 

(c) L-junctions given by the proposed approach;                      (d) Y-junctions given by the proposed approach;               

 

(e) X-junctions given by the proposed approach;                      (f)  all junctions given by the proposed approach;               

Junction characterization on a natural image containing textures. (a) shows the result of the Pj on gPb detector; (b) shows the result of Harris detector; (c)-(f)  display the L-, Y- and X- and all junctions given by the proposed approach. The location of each junction is indicated by a red cross and circles have a radius equal to the corresponding scale. The color of the junction indicates the significance (red for big significance and blue for small one).

 

References: 

  1. An accurate and contrast invariant junction detector. [PDF]
    Gui-Song Xia, Julie Delon, Yann Gousseau,
    The 21th International Conference on Pattern Recognition (ICPR): Tsukuba, Japan    , 2012.
  2. Accurate junction detection and characterization in natural images, (Technical Report). [PDF]
    Gui-Song Xia, Julie Delon, Yann Gousseau,
    Technical Report     HAL-00631609