BASIC MORPHOLOGICAL IMAGE PROCESSING OPERATIONS: A TUTORIAL

2. Mathematical Morphology (Serra, 1982)

Mathematical morphology is based on geometry. The theoretical foundations of morphological image processing lies in set theory and the mathematical theory of order. The basic idea is to probe an image with a template shape, which is called structuring element, to quantify the manner in which the structuring element fits within a given image.

2.1 Binary morphology

A binary image A in the N-dimensional Euclidean space E^N can be considered as a subset of E^N: A I E^N. The image foreground (set A) consists of all points zIE^N having value 1 and the image background (set A^C) consists of all points zIE^N having zero value.

The translation of A by a point x is denoted by A+x or (A)_x and is defined as follows:

(1)

Geometrically, A+x results by translating every point of A along the vector x. Figure 1 illustrates A+x and B+x, where A, B I E². Notice that if a point y of the input image A coincides with the origin, then this point in the translated image A+x corresponds to point x.

Figure 1. Translations of disk A and square B by x (A, B I E²). Notice that the centre of the disk which coincides with the origin, after the translation by x corresponds to point x.

The reflection of A is denoted by –A or A^s and is defined as follows:

(2)

Figure 2 illustrates -A, where A I E².

Figure 2. Reflection of rectangular A (AIE²).

Erosion

The fundamental operation of mathematical morphology is erosion. All mathematical morphology depends on this notion. The erosion of an input image A by a structuring element B, is defined as follows:

(3)

This means that in order to perform the erosion of A by B we translate B by x so that this lies inside A. The set of all points x satisfying this condition constitutes . Figure 3 illustrates the erosion of a triangle by a disk.

Figure 3. is the internal triangle, according to eqn (3).

The erosion of an image can also be found by intersecting all translates of the input image by the reflection of the structuring element:

(4)

An example of performing erosion using eqn (4) is illustrated in Figure 4, (A, B I E²). The arrows denote the origin and the shaded area represents the set points.

Figure 4. results from (the intersection of some translations of A), according to eqn (4).

Dilation

The dual operation to erosion is dilation. Dilation of an input image A by a structuring element B, is defined as follows:

(5)

This means that in order to perform the dilation of A by B we first translate B by all points of A. The union of these translations constitutes . Figure 5 illustrates the dilation of a triangle by a disk.

Figure 5. is the external triangle with rounded corners, according to eqn (5).

Dilation is both commutative and associative:

and (6)

By commutativity: (7)

This means that dilation can also be formed by translating the input image by all points in the structuring element and then taking the union. An application of eqn (7) is illustrated in the example of Figure 6, (A, BE²).

Figure 6. results from , according to eqn (7).

Basic Properties of Erosion and Dilation

(i) If the origin lies inside the structuring element, erosion is anti-extensive (shrinks the input image)

(8)

and dilation is extensive (expands the input image)

(9)

(ii) Both erosion and dilation are translation invariant relative to the input image:

and (10)

Since dilation is commutative, it is translation invariant relative to the structuring element. On the other hand erosion is not:

but (11)

(iii) Both erosion and dilation are monotonically increasing relative to a specific structuring element:

and (12)

Since dilation is commutative, it is monotonically increasing relative to a specific input image, but erosion is not:

but (13)

(iv) Erosion is dual to dilation (it is defined via dilation by set complementation) and vice versa:

or (14)

Thus, eroding an image can be accomplished by dilating the complement of the image, while dilating an image can be accomplished by eroding the complement of the image. In other words, dilation expands the image foreground and shrinks its background, whilst erosion shrinks the image foreground and expands its background.

Opening

A secondary operation of great importance in mathematical morphology is the opening operation. Opening of an input image A by a structuring element B is defined as follows:

(15)

An equivalent definition for opening is:

(16)

This means that in order to open A by B we first translate B by x so that this lies inside A. The union of these translations constitutes . For instance, the opening of a triangle A by a disk B (the origin coincides with the centre of the disk) is the triangle A with rounded corners. In general, opening by a disk rounds or eliminates all peaks extending into the image background.

If , then A is invariant under opening by B and it is called B-open.

Closing

The other important secondary operation is closing. Closing of an input image A by a structuring element B is defined as follows:

(17)

For instance, closing a triangle A by a disk B (the origin is on the centre of the disk) yields the same triangle A. In this case and we say that A is B-close. In general, closing by a disk rounds or eliminates all cavities extending into the image foreground.

Basic Properties of Opening and Closing

(i) Opening is anti-extensive and closing is extensive:

and (18)

(ii) Both opening and closing are translation invariant:

and (19)

(iii) Both opening and closing are monotonically increasing:

and (20)

(iv) Opening and closing are is dual operations:

and (21)

(v) Both opening and closing are idempotent:

and (22)

The importance of idempotency is that once an image has been opened (closed), successive openings (closings) do not alter the result (Haralick, et al., 1987). Idempotent transformations are of particular significance since they correspond to the ideal non-realisable band-pass filters of conventional linear filtering; once an image is ideally band-pass filtered, further filtering does not alter the result.

Open-Close Filters (Dougherty and Astola, 1994)

Suppose that is an image A corrupted by pepper noise N (i.e. granules of noise on the image background). If the corrupted image is opened by a suitable structuring element B, pepper noise is reduced or totally eliminated. It can be proven that:

(23)

If A is B-open , the previous relation becomes:

(24)

It is obvious from relations (23) and (24) that an ideal structuring element B has the proper shape and size, so that it completely restores the original image A; i.e. it eliminates the granules of pepper noise and it permits the whole image A to pass through the filter. If N is salt type noise (i.e. holes on the image foreground), we can reduce or eliminate this noise and restore the original image A by closing the corrupted image by a proper structuring element B.

Aiming at reducing or eliminating both, salt and pepper noise, we can first open the corrupted image by a proper structuring element B and then close the resulting image by the same structuring element. In this way we implement an open-close filter. In this case, the structuring element must be not only large enough to eliminate the noise granules but also small enough to fit between salt holes. If the structuring element does not fit between two salt holes, during opening the erosion by B will create larger holes, thus degrading the image foreground. If such a structuring element cannot be chosen, we can first use an open-close filter with a very small size structuring element, followed by an open-close filter with a larger size structuring element, which is followed by an open-close filter with an even larger size structuring element etc. In this way we implement an alternating sequential filter, which gradually eliminates noise components from smallest to largest.

2.2 Grey-scale morphology

A grey-scale image A in the N-dimensional Euclidean space is represented in an N+1-dimensional orthogonal system (N axes for spatial co-ordinates and one for grey-levels). For signals N=1 and for images N=2.

Let at any point z outside the signal (image) frame. We define:

Finite Domain of f(z): (25)

Spatial translation of f by x: (26)

Vertical translation of f by y (offset): (27)

Translation of f in the plane ( f is a signal): (28)

Reflection of g through the origin of spatial co-ordinates axes: g(-z) (29)

Erosion

The erosion of f by a structuring element g at a point x is defined as follows:

(30)

This means that we translate spatially g by x (so that its origin is located at point x) and then we find the minimum of all differences of values of f with the corresponding values of the translated g, .

An equivalent definition of erosion is:

(31)

Here we translate f by -x instead of translating g by x. The example of Figure 7 illustrates how we use eqns (30) and (31) to perform (fΘg)(x), (N=2).

Figure 7. f(1,2)=3, x=(1,2), z = {(0,0),(0,1),(-1,0),(-1,1)} and

Indeed, according to eqns (30) and (31) :

Dilation

The dilation of f by a structuring element g at a point x is defined as follows:

(32)

This means that we translate spatially the reflection of g, g (-z), so that its origin is located at point x and then we find the maximum of all sums of values of f with the corresponding values of the translated reflection of g, .

An equivalent definition of dilation is:

(33)

An example of performing dilation using eqns (32) and (33) is illustrated in Figure 8. Grey-scale erosion, dilation, opening and closing possess the properties of binary erosion, dilation, opening and closing, respectively.