Creating Color Image Features Based on Morphology Image Processing

Digital images are considered as one of the most widely used types of data since they are used in many applications and vital systems such as facial recognition systems, fingerprint recognition systems, and many others [1-4]. The digital image is characterized by the ease of processing because it is represented by a three-dimensional matrix (one dimension for each of the red, green, and blue colors) and as shown in Figure 1, (Instead of using color image with huge size, we can use the image features vector, which contains a fixed small number of elements extracted from the image using a method of features extraction).

1.png

Figure 1. Color image representation

Applications that use color digital images require a high speed of implementation, but the size of the digital image is usually very large, which may lead to a negative impact on the performance of the system that uses the digital color image. This in turn requires searching for alternatives. The most important of which is the use of the features vector to represent the image and as it is shown in Figure 1.

Image features vector is an array of a fixed number of values, and this array can be used as an identifier or key to identify the image. To use the features vectors in the recognition system, the features vector must satisfy the following requirements [5-14]:

2.png

Figure 2. Features extraction phase in the color image recognition system

3.png

Figure 3. Erosion example

• It should be unique for each image and not be repeated for more than one image.
• It must have a fixed size and a specified number of values.
• It must be digital and processable.
• The method of calculating should be easy and effective, and it does not take much time to extract it.

The stage of extracting the features of digital images and preserving these features is one of the important stages in building the discrimination system (see Figure 2) (extracted image features must be added to images features database, which will be used in a classification system, based for example on artificial neural network (ANN)), which in the future will affect its efficiency and the accuracy of reactions to this system [6, 7].

Color digital images are processed in a variety of ways. What we need in this research is the use of some simple operations, including the erosion process. The erosion process is considered as one of the most important basic digital image morphological operations, and it is implemented using a structuring element (SE) that is composed of an array of values of one or zero, where one indicates the entry of the neighbor into the executing process, while zero indicates the exclusion of the neighbor from the processing process [14, 15].

Erosion selects the minimum neighbor value from the neighbor's values which are included by the selected SE; Figure 3 shows an example of executing erosion operation (erosion uses the input image and the structuring elements and it returns the minimum neighbor value which is cover by the structuring element) [16-23].

The organization of this paper is as follows. Section 2 presents K-means clustering. Section 3 demonstrates the proposed approach. Implementation and experimental results are conducted in Section 4, followed by result analysis in Section 5. Section 6 provides the conclusions of this research paper.

The aim of this research is to provide a simple and easy to implement method for extracting the properties of any color digital image to reduce as much as possible the extraction time and retrieve unique properties for each image. So, those properties can be used as an identifier or key to the image.

Hit and miss operation (HAMO) is one of the powerful morphological operations for finding objects and their locations in a digital image. HAMO can be defined entirely in terms of erosion only. HAMO is very useful for detecting specific objects that are intended to extract such as isolated points, two connected points, three connected points, crosses, squares, triangles, ridges, corners, junctions, and so on [24-28].

HAMO can be implemented using erosion operation. Erosion is a pixel operation which returns the minimum value from the neighbor's values which are covered by a structuring element, and this element contains zeros and ones; one means that the associated neighbor is covered as shown in the following example (see Figure 6).

6.png

Figure 6. Erosion operation

To detect any object in the color image we must follow the following steps:

● Select a structuring (SE) element which is equal to the object to be detected (see Figure 7).

7a.png

Figure 7. Detecting an object equal to SE1

7b.png

Figure 8. Eroding the image with SE1

7v.png

Figure 9. Finding the complement of the image

7d.png

Figure 10. Eroding the complement of the image with SE2

● Erode the image with SE1 (see Figure 8).

● Find the complement of the image (see Figure 9).

● Erode the complement of the image with SE2 (see Figure 10).

● Apply ANDing b and d (see Figure 11).

11.png

Figure 11. ANDing b and d

12.png

Figure 12. Counting the appearance of SE in the image

The proposed method can be implemented applying the following sequence of operations (see Figure 13).

13.png

Figure 13. Proposed method algorithm

● Get the color image.

● Convert the color image to gray.

● Convert the gray image to binary image, Figure 14 shows a sample output of implementing these steps.

● Defining the required structuring elements needed to count the associated objects in the binary image, those objects are: isolated point with black color, isolated point with white color, two connected points, and three connected points, the structuring elements of the object are shown in Figures 14 to 17.

● Apply hit and miss operations using each structuring element: one time to count isolated points with black color, one time to count isolated points with white color, 4 times to count two connected points, two times to count three connected points.

● For each object apply summations of the obtained matrices to get the counts of the objects, this count can be used as a color image feature.

● Figure 18 illustrates the process of counting the object 3 connected points in an image.

14.png

Figure 14. Counting the appearance of SE in the image

15.png

Figure 15. SE to detectisolated points

16.png

Figure 16. SEs to detect two connected points

17.png

Figure 17. SEs to detect three connected points

18.png

Figure 18. Counting 3 connected points

From the obtained experimental results, we can see the following facts:

● K-means clustering method provides a good ability to form image features vectors, each of them satisfies the requirements.

● We cannot predict the extraction time in K-means clustering, because it depends on the number of executed passes and the image size.

● Resizing the image to a smaller size reduces the extraction time in K-means clustering and in the proposed method.

● The extraction time in the proposed method depends on the image size, and reducing the image size by resizing will reduce the extraction time.

● The proposed method provides a unique features vector for each image.

The proposed method speeds up the process of features extraction, thus enhancing the performance of the digital image features extraction process; this can be seen in Tables 7 and 8.

Table 7. Speedup of the proposed method using source images

Table 8. Speedup of the proposed method using resized images