You are on page 1of 7

Practical Course 3. Image Filtering 15.12.

2015
Adalbert Zielonka 3087565
Julio Correa
3087974
Yannik Blum
4019792

Task 1: Image sharpening (filtering of angiographic images).


Introduction:
Different values in the unsharp mask filter using the image DVI_PR_4.TIF have been applied
to choose the best values where the filter improved the recognisability of the blood vessels.
Methods:
First the unsharp mask filter was applied in different combinations of Gaussian radius (G) and
mask weight (M) shown in Fig1.1 and Fig1.2.The best values for the unsharp mask filter were
chosen by analyzing the changes in the images. The unsharp mask is generated by subtracting
the original image with a blurred image of itself as shown in eq.1, when applying the filter the
mask is added as in eq.2, where S is a scaling constant.
(1) (, ) = (, ) (, )
(2) (, ) = (, ) + (, )
Using images "DVI_BR_1.TIF and "DVI_BR_2.TIF" the angiographic difference image
analysis was determined, the unsharp mask filter was applied with the chosen values before the
subtraction and for comparison without using the filter, and for the two cases the linear contrast
stretching was applied at the end.
Results:

Fig.1.1- In this image different values of the


Gaussian radius were applied while the Mask
weight values remained the same ( = .20).
In A) The original image, B) = 5, C) = 10
and D) = 15.

Fig.1.2- In this image different values of the


Mask Weight were applied while the Gaussian
radius values remained the same ( = 5).
In A) The original image, B) = .30, C) =
.60 and D) = .90

Fig.1.3- In this image A) the original image, B) the choice for the unsharp filter mask ( = 10, =
0.4), C) the values where image is too smoothed ( = 15, = 0.2) and D) too much noise on the
image ( = 2, = 0.9).
When applying different values of Gaussian radius shown in Fig.1.1, it can be seen that it affects the
smoothing or unsharping of the image, also increasing the value increases the contrast, while the mask
weight as shown in Fig.1.2. determines the weight of filtering and increasing its value provides
additional edge enhancement.

Fig.1.4. Subtraction of images (with the unsharp mask filter on LEFT and without the filter on RIGHT).

Conclusion and discussion


The unsharp mask filter can be used for enhancing edges and thus sharpening the image for better
visualization although sometimes noise can affect the image, this filter can compensate this noise
effect, but also it was noticed that increasing the weight mask value too much could provoke noise
enhancement like in Fig.1.3.B. When comparing the images in Fig1.4. it can be seen that when the filter
is applied the result is a better sharpened image than when the filter was not used and by that, giving
more details on the edges and improving image contrast.

Task 2. Smoothing filter (filters for noise suppression).


Introduction
For this task an image of a radiographic test sample shown in Fig.2.1.was taken under noisy
conditions and different kernel sizes with a mean and Gaussian blur filters have been tested to
reduce the noise affecting the image.
Method
The Mean and Gaussian blur filter were applied with different kernel sizes for reducing the
noise and by so smoothing the image, for determining how the noise was affected, the histogram
in each case was taken in a ROI of 80x80 in a homogeneous area, and the mean and standard
deviation values were taken in account for comparing the reduction of noise.
Results

Fig.2.1.Original Image with histogram of the ROI.

Fig.2.2. Application of mean filter with 3x3 Moving Average with Radius: 1px.

Fig.2.3. Application of mean filter with 5x5 Moving Average with Radius: 2px.

Fig.2.4. Application of mean filter with 7x7 Moving Average with Radius: 3px.

Fig.2.5. Application of Gaussian blur filter with 7x7 Gaussian Lowpass with Radius: 3px.

Original Image
3x3 Moving Average
5x5 Moving Average
7x7 Moving Average
7x7 Gaussian Lowpass

Mean
183.915
183.881
183.881
183.901
183.926

StdDev
15.295
7.591
5.507
4.349
5.442

Fig.2.6. Table made from the mean and standard deviation values corresponding to the original
image in Fig.2.1 and filtered images shown from Fig.2.2. to Fig.2.5.
As shown in Fig.2.2 through Fig.2.5 while increasing the size of the kernel the smoothing of the image
is increased. In the table of Fig.2.6. The values of the histogram in each case are shown, while increasing
the size the value of the standard deviation decreases. In this case the mean filter reduced more noise
than the Gaussian blur filter.

Discussion and conclusions


With applying low pass filters noise can be reduced and the image is smoothed, if the kernel size
increases the smoothing increases, although by means of computation demand it also increases, the noise
reduction is a good factor but the blurring of the image is affected and thus some details in the image
can be lost.

Task 3: Further medical applications of image filtering


Image Enhancement in Digital Radiography.
Introduction
The goal of this task is to test different parameters in the laplacian kernel for enhancing the image
DVI_BR_2.TIF with the laplacian operator and to compare results with the unsharp mask filter.

Method
As shown in eq.1 the values have been modified in the laplacian kernel with diagonal components (2)
in order to choose the best parameter for X for enhancing the image, if we relate (1) and (2) the A
1
value in imageJ is simplified and by this = 8 , so values bigger than 8 have been chosen, different
visualizations of changing the parameter in the kernel are shown from Fig. 3.2 to Fig.3.6. The unsharp
mask filter for comparing was applied in Fig.3.7.

(1)

(2)
Results
The following results were obtained using the convolve tool with the normalize kernel value checked,
this for preserving the image brightness, as shown in Fig.3.2. Beginning with a parameter X=8.5 didnt
result in a very good enhancement while a good enhancement was obtained using a parameter X=10.
In Fig 3.7. The unsharp mask filter was applied with a Gaussian radius of 10 and a mask weight of .4.

Figure 3.2. Image with + = . .

Figure 3.3. Image with + = .

Figure 3.5. Image with + = 10.

Figure 3.6. Image with + = 10.5.

Figure 3.7. Image filtered with unsharp mask, values


= , =.

Discussion and Conclusion


The application of the laplacian operator is used for enhancement of the image, this method can be
used to find fine details in an image. A disadvantage is with the presence of noise as it can be amplified
by the filter and thus making the image more noisy. For comparing the application of the enhancement
in both filters the laplacian and the unsharp mask similar results were shown, although details in the
image were appreciated better with the laplacian filter, depending on the approach of the medical
application the different kinds of filters must be implemented also considering the noise or other
artefacts for choosing the proper filter.

You might also like