Asplünd’s metric defined in the Logarithmic Image Processing (LIP) framework for colour and multivariate images

Asplünd’s metric defined in the Logarithmic Image Processing (LIP) framework for colour and multivariate images

Abstract

Asplünd’s metric, which is useful for pattern matching, consists in a double-sided probing, i.e. the over-graph and the sub-graph of a function are probed jointly. It has previously been defined for grey-scale images using the Logarithmic Image Processing (LIP) framework. LIP is a non-linear model to perform operations between images while being consistent with the human visual system. Our contribution consists in extending the Asplünd’s metric to colour and multivariate images using the LIP framework. Asplünd’s metric is insensitive to lighting variations and we propose a colour variant which is robust to noise.

Asplünd’s metric defined in the Logarithmic Image Processing (LIP) framework for colour and multivariate images

Guillaume Noyel, Member IEEE
International Prevention Research Institute
95 cours Lafayette
69006 Lyon, France
                  
Michel Jourlin
Laboratoire Hubert Curien, UMR CNRS 5516
18 rue du Professeur Benoît Lauras
42000 Saint-Etienne, France


Index Terms—  Asplünd’s distance, colour and multivariate images, Logarithmic Image Processing, double-sided probing, pattern recognition

1 Introduction

The LIP framework has been originally defined by Jourlin et al. in [1, 2, 3]. It consists in performing operations between images such as addition, subtraction or multiplication by a scalar with a result staying in the bounded domain of the images, for example for grey-scale images on 8 bits. Due to the relation between LIP operations and the physical transmittance, the model is perfectly suited for images acquired by transmitted light (i.e. when the observed object is located between the source and the sensor). Furthermore, the demonstration, by Brailean [4] of the compatibility of the LIP model with human vision has enlarged its application for images acquired in reflected light.

The Asplünd’s metric initially defined for binary shapes [5, 6] has been extended to grey-scale images by Jourlin et al. using the LIP framework [7, 8]. One of the main property of this metric is to be strongly independent of lighting variations. It consists in probing a function by two homothetic functions of a template, i.e. the probe. As the homothetic functions are computed by a LIP multiplication, the distance is consistent with the human vision.

After a reminder of the LIP model and the Asplünd’s metric for grey-scale images, a definition is given for colour images in order to perform colour matching. A variant of the metric, robust to the noise, is also proposed. Examples will illustrate the properties of the metric.

2 Prerequisites

2.1 LIP model

Given a spatial support , a grey-scale image is a function with values in the grey-scale :
.

In the LIP context, corresponds to the “white” extremity of the grey-scale, which means to the source intensity, i.e. when no obstacle (object) is placed between the source and the sensor. Thanks to this grey-scale inversion, will appear as the neutral element of the logarithmic addition. The other extremity is a limit situation where no element of the source is transmitted (black value).This value is excluded of the scale, and when working with 8-bit digitized images, the 256 grey-levels correspond to the interval of integers .

Due to the relation between the LIP model and the transmittance law, [2], the addition of two images and corresponds to the superposition of the obstacles (objects) generating respectively and . The resulting image will be noted:

From this law, the multiplication of an image by a positive real number is defined by:

Physical interpretation [2]: In the case of transmitted light, the sum consists in stacking twice the semi-transparent object corresponding to . Therefore, the LIP multiplication of by a scalar corresponds to changing the thickness of the observed object in the ratio . If , the thickness is increased and the image becomes darker than , while if , the thickness is decreased and the image becomes brighter than .

2.2 Asplünd’s metric for grey-scale images

Definition 1.

Given two images and defined on , is chosen as a probing function for example, and we define the two numbers: and . The corresponding “functional Asplünd’s metric” is:

(1)

Physical interpretation [8]: As Asplünd’s metric is based on LIP multiplication, this metric is particularly insensitive to lighting variations, as long as such variations may be modelled by thickness changing.

In order to find in an image the location of a given template, the metric can be adapted to local processing. The template corresponds to an image defined on a spatial support . For each point of , the distance is computed on the neighbourhood centred in , with being the restriction of to .

3 Asplünd’s metric for colour and multivariate images

A colour image , defined on a domain , with values in , , is written:

(2)

with , , being the red, green and blue channels of , and being a vector-pixel.

A colour image is a particular case of a multivariate image which is defined as , with being an integer number corresponding to the number of channels [9].

Definition 2.

Given two colours , , their Asplünd’s distance is equal to:

(3)

with and

Strictly speaking, is a metric if the colours are replaced by their equivalence classes .

Colour metrics between two colour images and may be defined as the sum ( metric) or the supremum () of on the considered region of interest :

(4)
(5)

with the cardinal of . In the same way:

Definition 3.

The global colour Asplünd’s metric between two colour images and on a region is

(6)

with

and
.

In figure 1, the Asplünd’s metric has been tested between the colour probe and the colour function on their definition domain . In this case, the Asplünd’s distance is equal to . The lower bound and the upper bound are determined as shown in Figure 1 (c).

(a) Colour function (b) Colour probe (c) Lower () and upper () bounds
Fig. 1: Computation of the Asplünd’s distance between two colour functions . Each colour channel of the function is represented by a line having the corresponding colour.

As for grey-scale images, the metric may be adapted to local processing. The template, (i.e. the probe) corresponds to a colour image defined on a spatial support . For each point of , the distance is computed on the neighbourhood centred in , with being the restriction of to .

Definition 4.

Given a colour image defined on into , , and a colour probe defined on into , , the map of Asplünd’s distances is:

(7)

with the neighbourhood corresponding to centred in .

In figure 2, the map of Asplünd’s distances is computed between a colour function and a colour probe. The minima of the map corresponds to the location of a pattern similar to the probe.

(a) Colour function (b) Colour probe (c) Map of Asplünd’s distance
Fig. 2: (c) Map of the Asplünd’s distance between a colour function and a colour probe. (a) and (b) Each colour channel is represented by a line having the corresponding colour.

As explained in [8], Asplünd’s distance is sensitive to noise because the probe lays on regional extrema produced by noise (Figure 1). In [8], Jourlin et al. have introduced a new definition of Asplünd’s distance with a tolerance on the extrema corresponding to noise. In this paper, we extend this definition for colour and multivariate images.

To reduce the sensitivity of Asplünd’s distance to the noise, there exists a metric defined in the context of “Measure Theory”. It will be called Measure metric or M-metric. Only a short recall of this theory adapted to the context of colour images defined on a subset is presented. Given a measure on , a colour image and a metric on the space of colour images, a neighbourhood of function may be defined thanks to and two arbitrary small positive real numbers and according to:

It means that the measure of the set of points , where exceeds the tolerance , satisfies another tolerance . This can be simplified, in the context of Asplünd’s metric:

  • the image being digitized, the number of pixels lying in is finite, therefore the “measure” of a subset of is linked to the cardinal of this subset, for example the percentage of its elements related to (or a region of interest ). In this case, we are looking for a subset of , such that and are neighbours for Asplünd’s metric and the complementary set of related to is small sized when compared to . This last condition can be written as:

    where represents an acceptable percentage and the number of elements in .

  • Therefore, the neighbourhood is

    (8)

We follow the same approach already used in [8] to compute the Asplünd’s distance with a tolerance . In figure 3, a tolerance of is used and consists in discarding two points. Therefore, the Asplünd’s distance is decreased from to .

(a) Colour function (b) Colour probe (c) Lower and upper bounds,
Fig. 3: Colour Asplünd’s distance with a tolerance of . (, ) are the scalars multiplying the probe without tolerance. (, ) are the scalars multiplying the probe with tolerance.

With this distance, a map of Asplünd’s distances can be defined.

Definition 5.

Given a colour image defined on into , , and a colour probe defined on into , , a tolerance , the map of Asplünd’s distances with a tolerance is:

(9)

with the neighbourhood corresponding to centred in .

After having introduced the colour Asplünd’s distance, examples are given.

4 Examples and applications

In figure 4, our aim is to find bricks of homogeneous colour inside a colour image of a brick wall. In the map of Asplünd’s distances of the image without noise , the regional minima correspond to the location where the probe is similar to the image (according to Asplünd’s distance). In the noisy image , the map of Asplünd’s distance is very sensitive to noise (fig. 4 d). Therefore, it is necessary to introduce a tolerance in the map , to find the minima corresponding to the bricks. One can notice that:

  • the minima are preserved into the map with a tolerance (fig. 4 c) compared to the map without (fig. 4 f)

  • the maps are insensitive to a vertical lighting drift.

The minima can be extracted using standard mathematical morphology operations [10, 11].

In figure 5, two images of the same scene, a bright image and a dark image , are acquired with two different exposure time. The probe is extracted in the bright image and used to compute the map of Asplünd’s distance in the darker image. By finding the minima of the map, the balls are detected and their contours are added to the image of figure 5 (b). One can notice that the Asplünd’s distance is very robust to lighting variations.

(a) Image and probe (b) Noisy image (c) Map (d) Map (e) Colour probe magnified (f) Map
Fig. 4: Maps of Asplünd’s distances without tolerance and with tolerance . In , a Gaussian white noise with zero-mean, variance 2.6 and spatial density has been used.
(a) Initial image and probe (b) Dark image Balls detected (c) Map
Fig. 5: Detection of coloured balls on a dark image with a probe extracted in the bright image . (a) The border of the probe is coloured in white.

5 Conclusion and perspectives

An Asplünd’s distance for colour and multivariate images has been introduced. Based on a double-sided probing of a function, this distance is particularly insensitive to the lighting variations. Moreover an alternative definition of the colour Asplünd’s distance robust to noise has been introduced. An example illustrates the robustness of the method to the lighting variations and to the noise. With this new distance, efficient colour or multivariate pattern matching can be performed in images with all the properties described in [8]. In future works, we plan to present the definition of a colour Asplünd’s distance with a colour LIP model [3] (already defined). We are also going to study the links between Asplünd’s probing and mathematical morphology.

References

  • [1] M. Jourlin and J.C. Pinoli, “A model for logarithmic image processing,” Journal of Microscopy, vol. 149, no. 1, pp. 21–35, 1988.
  • [2] M. Jourlin and J.C. Pinoli, “Logarithmic image processing: The mathematical and physical framework for the representation and processing of transmitted images,” in Advances in Imaging and Electron Physics, Peter W. Hawkes, Ed., vol. 115, pp. 129 – 196. Elsevier, 2001.
  • [3] M. Jourlin, J. Breugnot, F. Itthirad, M. Bouabdellah, and B. Closs, “Chapter 2 - Logarithmic image processing for color images,” in Advances in Imaging and Electron Physics, Peter W. Hawkes, Ed., vol. 168, pp. 65 – 107. Elsevier, 2011.
  • [4] J.C. Brailean, B. Sullivan, C.T. Chen, and M.L. Giger, “Evaluating the EM algorithm for image processing using a human visual fidelity criterion,” in Acoustics, Speech, and Signal Processing, 1991. ICASSP-91., 1991 International Conference on, Apr 1991, pp. 2957–2960 vol.4.
  • [5] E. Asplünd, “Comparison between plane symmetric convex bodies and parallelograms,” Mathematica Scandinavica, vol. 8, pp. 171–180, 1960.
  • [6] B. Grünbaum, “Measures of symmetry for convex sets,” in Proceedings of Symposia in Pure Mathematics, 1963, pp. 233–270 vol.7.
  • [7] M. Jourlin, M. Carré, J. Breugnot, and M. Bouabdellah, “Chapter 7 - Logarithmic image processing: Additive contrast, multiplicative contrast, and associated metrics,” in Advances in Imaging and Electron Physics, Peter W. Hawkes, Ed., vol. 171, pp. 357 – 406. Elsevier, 2012.
  • [8] M. Jourlin, E. Couka, B. Abdallah, J. Corvo, and J. Breugnot, “Asplünd’s metric defined in the logarithmic image processing (LIP) framework: A new way to perform double-sided image probing for non-linear grayscale pattern matching,” Pattern Recognition, vol. 47, no. 9, pp. 2908 – 2924, 2014.
  • [9] G. Noyel, J. Angulo, and D. Jeulin, “Morphological segmentation of hyperspectral images,” Image Analysis & Stereology, vol. 26, no. 3, 2007.
  • [10] G. Matheron, Eléments pour une théorie des milieux poreux, Masson, Paris, 1967.
  • [11] J. Serra and N.A.C. Cressie, Image analysis and mathematical morphology: Vol. 1, Academic Press, London, 1982.
Comments 0
Request Comment
You are adding the first comment!
How to quickly get a good reply:
  • Give credit where it’s due by listing out the positive aspects of a paper before getting into which changes should be made.
  • Be specific in your critique, and provide supporting evidence with appropriate references to substantiate general statements.
  • Your comment should inspire ideas to flow and help the author improves the paper.

The better we are at sharing our knowledge with each other, the faster we move forward.
""
The feedback must be of minimum 40 characters and the title a minimum of 5 characters
   
Add comment
Cancel
Loading ...
103393
This is a comment super asjknd jkasnjk adsnkj
Upvote
Downvote
""
The feedback must be of minumum 40 characters
The feedback must be of minumum 40 characters
Submit
Cancel

You are asking your first question!
How to quickly get a good answer:
  • Keep your question short and to the point
  • Check for grammar or spelling errors.
  • Phrase it like a question
Test
Test description