Project Report

Computer vision

Submitted by:

Abhishek Singh 401703002

Mallika Pushkarna 401703013
Pranav Kakkar 401703019
Arantish Singh 401883001

Computer Science and Engineering Department

TIET, Patiala

December 2019
 A novel binary feature descriptor to discriminate normal
and abnormal chest CT images

Task 1
Identify type of noise?

The noise throughout most of these images is commonly referred to as salt and pepper
noise this type of noise consists of certain amounts of the pixels in the image either being
black or white, the most common method of removing such noise is known to be through a
medium filter “ filtering is less sensitive than linear techniques to extreme changes in pixel
values, it can remove salt and pepper noise without significantly reducing the sharpness of
an image” Therefore the approach of using a medium filter within this assignment was
applied

Input, Grayscale and Noise removal

The problem with the initial image is the amount of noise present is very high which needs to
be removed before any other processes can be applied to find the circles within the image,
noise can be seen in the screenshot above and causes distortion within the image in terms of
brightness and darkness “Noise which varies randomly above and below a nominal
brightness value” (Bovik, 2009) Therefore the first step to providing the solution to this
problem is the removal of any noise within the images inputted into the program. The first
step to this task was to read the image into the program “Selected_Image = imread
(image_jpg);” This allows for the desired input images to be read, after the image is read it is
then converted from RGB to grayscale using the rgb2gray function, this allows for the
removal of hue and saturated information while allowing for the ability to keep the images
desired luminance.

Once the image has been read and converted the noise then can be removed. The noise is
removed using the medium filtering, this is done through a separate function called
“customfilter” The function applies a medium filter to the grayscale image, which then allows
for each outputted pixel to contain a medium value in the 3-by-3 region around the
corresponding pixels within the image. The medium filter follows a similar approach to that
of smoothing techniques, the technique of using corresponding pixels allows for the better
ability of the removal of noise without reducing the sharpness of an image therefore making
this the appropriate function to reduce the salt and pepper noise within the images for this
assignment.
After the noise has removed the image is then implemented into a subplot to output the
image with its corresponding input image, this is done throughout the project to all later
images implemented within the project, an example of the noise removal and subplot is
shown above.
Task 2
Converting the Greyscale Image to a Binary Image

The next key step to solving the problem is to convert the current greyscale image into a
binary image. The reason behind converting the image into a binary image is it allows for
the better ability to detect objects within an image (circles) the reason behind this is due to
the fact now only two different colours exist within the image (black, white).The image is
converted into a binary image by using the im2bw function “binary_picture =
im2bw(median_filtering_Image, 0.2);” This function uses the filtered image as an input and
then converts the pixels within the image dependent upon the luminance level of each pixel
to either the value 1 (white) or the value 0 (black).

Create a morphological structuring element (STREL)

This step involves creating a structuring element in the form of a rectangle “se1 = strel('disk',
2);” This specific function creates a flat disk shape structure where “2” specifies the radius.
This is done twice to create two separate structures which are applied to the initial binary
image, creating two separate images “postOpenImage_1” and “postOpenImage_2” Once
these structures have been applied the first image is applied into the subplot and shown to
the user while the other is saved for later image processing to help solve to problem in later
stages.
 Inversion of the Opened Image
This step involves the inversion of the binary image which is completed by simply using a
ones function, this function basically creates an array of all ones, this is implemented to the
binary image “inverted = ones(size(binary_picture));” Once this is done the following is
implemented
“invertedImage_1 = inverted - postOpenImage_1;” This line basically takes the inverted
image with the array of ones and takes the Open image from this which in turn inverts the
image by swapping all previous 0 pixels to 1 pixels and 1 pixels to 0, creating an inverted
image as shown above. This is completed to allow for segmentation to be completed in the
following step to allow for the program to better find the circles within the image.

Creating an initial Contour and implementing segmentation

This step initially involves creating and specifying the initial contour “mask = zeros (size
(invertedImage_1)); mask (50: end-50, 50:end-50) = 1;” The mask is a binary image that
specifies the initial state of the active contour, this involves the boundaries of the images
regions in this case the white background and black regions outside the main part of the
Image, this all allows for contour evolution to occur to allow for segmentation of the image
to occur. Once the image is segmented it will produce a more clear way of finding the
circles within the lung as this means the foreground and background no longer cause any
issues when attempting to find the circles as seen above in the screenshot.

Task 3
Combination Image and Find Circles

The next process required is to find the circles within the lung,
within this program the image is created using a combination of
the inverted image and contour segmentation “mix_Image_1 =
invertedImage_1 + bw_1;” This creates an initial image that the
circles can be found upon, from this the image is then converted
into black and white using “im2bw” then the final image uses the
medium filter “medfilt2” to further filter the image to allow for
the best possible ability to find any circles within the lung. The
main process within this step is now to find the circles this is
done using the imfindcircles function
“[centers,radii] = imfindcircles(final_1,[1
9],'ObjectPolarity','dark','Sensitivity',0.88); This function looks for dark sensitivity within
the above image using the parameters of centres and radius with a sensitivity specification,
 from this the circles are then displayed using green circles
“viscircles(centers,radii,'EdgeColor','g'); “ After this is completed the image is then
displayed as seen above in the screenshot with the circles all being highlighted in green.
The circles are also counted “display(size(centers, 1), ' Numbers of Circles');” this allows
for all circles found to have a centre to be counted and outputted into the command
window.

Task 4
Segment Circles
After the circles have been counted and found the circles need to be Segmented and then
detected. This is done by firstly creating a segmented picture “segment_pic = final_2 -
final_1;” This is done by taking a picture without the circles located to a picture with the
circles located, therefore leading to an image with the circles totally segmented away from
the picture allowing for a precise way to find the boundaries of each circle “[B] = (pre
bwboundaries _colour_pic,'holes');“ This code basically finds all circles within the image and
the boundaries and then colours each circle with a fill of green to identify they have been
found as seen in the screenshot above. Therefore solving all problems indicated within the
brief as the circles have been counted found and all noise has been removed.

References
Bovik, A. (2009) The essential guide to image processing. [Online] Available from:
http://links.uwaterloo.ca/amath391w13docs/bovik_image_processing.pdf [Accessed
13/10/2019]
Matlab (2019) Remove Salt and Pepper Noise from Images. [Online]
Available from: http://uk.mathworks.com/help/vision/ug/remove-salt-and-
pepper-noise-from- images.html?refresh=true [Accessed 13/10/2019]
 DECLARATION

We hereby declare that the design principles and working prototype model of the project entitled
A novel binary feature descriptor to discriminate normal and abnormal chest CT images is an
authentic record of our own work carried out in the Computer Science and Engineering
Department, TIET, Patiala, under the guidance of Dr.Shailendra Tiwari during 5th semester
(2019).

ACKNOWLEDGEMENT

We would like to express our thanks to our mentor Dr. Shailendra Tiwari. He has been of
great help in our venture, and an indispensable resource of technical knowledge. He is truly an
amazing mentor to have.

We are thankful to Shailendra sir and our friends who devoted their valuable time and helped
us in all possible ways towards successful completion of this project. We thank all those who
have contributed either directly or indirectly towards this project.

Lastly, we would also like to thank our families for their unyielding love and encouragement.
They always wanted the best for us and we admire their determination and sacrifice.