Similarities to the Opening Method

3.2.1 Similarities to the Opening Method

The CT provided by the doctors were patients lying on a platform. This can be seen in figure 3.2(a). This would alternate the results. So we would have to eliminate the platform, so that an accurate result would be gained. All images were unable to be conducted so only one patients image were provided to show the result.

We used a technique similar to the opening [28], first using several erosion than the method several dilation to eliminate the platform. Before using the technique similar to opening, a threshold of 150 was used to use the pixel that has a smaller value as a background. This would allow pixels values that are smaller or other outlier pollutants to be ignored. This would speed up the proposed technique and getting a faster result. This paper used N4 as structuring element to be tested.

(a) (b)

Figure 3.2 (a) Typical chest CT (b) Platform used in image capture (Source:http://en.wikipedia.org/wiki/File:64_slice_scanner.JPG)

1

1 1 1

1

Figure 3.3 The standard structuring elements (SE) N₄

Table 3.1 A simple dilation and erosion rule of using N4 as SE

Operation Rule Dilation Easily said the output is the input of pixel neighboring (up, down, left,

right and itself) of the biggest pixel value which replaces the original pixel. If binary image, the neighboring points, if a pixel is not the background, if it isn’t zero, the operated pixel will be replaced by 1.

Erosion Easily said the output is the input of pixel neighboring (up, down, left, right and itself) is the smallest pixel value which replaces the original pixel. If binary image, the neighboring points, if a pixel is the background, if it is zero, the operated pixel will be replaced by 0.

23

24

According to binary image’s dilation, as seen in Table 3.1, this provides a prompt discussion. The three tables in Figure 3.4 show the dilation process. We use a pixel for this discussion. This pixel is marked by arrow heads. In Figure 3.4(a) we can see this pixel is neighbored by green marked pixels. The red marked is neighbored by foreground pixel, thus the pixel that is being treated (marked red) will be transformed into a foreground pixel. In Figure 3.4(b) after treating all the pixels in the image, the results can be seen in 3.4(c).

According to gray level imaging we will use Figure 3.5. We use a pixel for this discussion. This pixel is marked by black arrow heads. In Figure 3.5(a), in the table seen above, the pixel’s value will be overtaken by the biggest value in its neighboring pixels as seen in Figure 3.5(b). In Figure 3.5(c) are the results after treating all the pixels in the image.

Next, we will be discussing about erosion. Figure 3.6 shows erosion in the binary image, the black arrow head shows the pixel. The neighboring pixels include background pixels, so the black arrow head pixels will be background pixels as seen in Figure 3.6(b).

The results can be seen in Figure 3.6(c).

According to the erosion on gray level image, the black arrow head show the pixel will be overtaken by the smallest pixel value in the neighboring pixels. As seen in Figure 3.7(b). The results can be seen in Figure 3.7(c).

(a) (b)

25

(c)

Figure 3.4 The dilation process for a binary image:

(a) The original binary image, according to the arrow heads pixel for dilation process.

(b) As the neighboring pixel is foreground pixel, so the value 0 is overtaken by 1.

(a) (b)

26

(c)

Figure 3.5 The dilation process for a grayscale image:

(a) The original gray level image, according to the arrow head pixel for a dilation process.

(b) As the neighboring pixel value, the biggest value will overtake the arrow head pixel.

(c) Results.

(a) (b)

27

(c)

Figure 3.6 The erosion process for a binary image:

(a) The original binary image, the arrow head pixel is being used in an erosion process.

(b) As the neighboring pixel is a background one, 1 will be overtaken by 0.

(c) Results.

1 0 0 1

1 1 1 0 0

0 0 1 0 0

0 0 0 1 0 0 0 0 1 0 0 0 1 0 0

0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0

(a) (b)

28

(c)

Figure 3.7 The erosion process for a grayscale image:

(a) The original grayscale image, the pixel marked by the arrow heads is being used in the erosion process.

(b) According to the neighboring pixel value, the smallest value will be overtaken.

(c) Results.

191 134 164 122 3

21 23 187 4 29

10 9 11 133 27

69 35 17 144 21

42 78 222 61 73

134 187 122 187

21 23 122 3 3 10 9 4 4 3 9 9 9 4 21 10 9 11 17 21 42 35 17 61 21

Due to the erosion process was not conducted thoroughly the platform in the image would not be able to be deleted. However, if the erosion process was conducted was conducted over thoroughly, after dilation the details in the image will still is lost. How many times we repeat the process is a topic that can be further discussed. After testing numerous times, and the optimal number of process is 13 times. This means doing a process of erosions 13 times and straight after doing the process of dilation 13 times.

The results can be seen in Figure 3.8.

(a) (b)

Figure 3.8 Delete the platform (a) Original patient 1 chest ct (b) Result of our method

In Figure 3.8 we can see the original patient 1 CT and the results of the method similar to opening. The experimental images in this chapter is all according to patient 1 unless noted otherwise.

29

3.2.2 Sobel Edge Detection

? g≥T

Figure 3.9 The flow chart for Sobel algorithm

Eliminating the platform, thus will leave the parts where will be processed, this is also the body.

The flow chart for Sobel algorithm is shown above. This is a edge detection methos which uses the calculation of a gradient, and we choose a threshold to see if the gradient is bigger than the threshold. This will show if this is an edge point.

Before this we will need a Threshold, T’s value to determine a Sobel algorithm’s process result. After numerous experiments on a number of T values, as seen in Figure 3.10, we have decided to use the T value of 100000. This value might be a big threshold, however, we don’t need all the detail edge in the image. So the Sobel process shouldn’t be conducted too sensitive.

..

30

(a) (b)

(c)

Figure 3.10 Use Sobel detection on different theshold,T (a) T=0 (b) T=10000 (c)T=100000

31