You are on page 1of 10

Computer Vision Project

By: Adam Puspabhuana


Lecturer : Rila Mandala , Phd

Subject: Computer Vision

Page 1
Outline :
• References
• Introduction
• Literature Review
• Research and Design Metodology
• Results and Discussion
• Final Proposed System for Random Data Set
• Conclusion and Future Work

Adam Puspabhuana,S.Kom – 3rd Semester, April 2015 Page 2


References
•http://opencv.org/
•http://www.digitalinferno.com/wiki/Wiki.jsp?page=Ja
vaCv
•https://code.google.com/p/javacv/
•https://github.com/bytedeco/javacv
•http://en.wikipedia.org/wiki/OpenCV

Adam Puspabhuana,S.Kom – 1st Semester, September 2014 Page 3


2nd Semester Mid Asignment (Group)
- This assignment is done by group (Adam,Amril and Iman)
- The purpose of this assignment is doing image detection with
opencv and java, which is tested using Netbeans 7.3, however we
have to make our own classifier (not using opencv classifer file) xml
file
-In order to have our own classifier, it is supported by opencv tools
such as opencv_training cascade,opencv_create samples, opencv
haartraining,and haartrainig stuff, you can see at this link
https://www.youtube.com/watch?v=KFC6jxBKtBQ to implement and
proof it.

Adam Puspabhuana,S.Kom – 1st Semester, September 2014 Page 4


2nd Semester Mid Asignment (Group)
- We collect all positive images means that all the images that we
want to train e.g car and we also collect negative images means that
all the images which is not related with positive images (car). All the
positive and negative images have to be in the same size of
picture(pixel) , make sure in grayscale images.
- We made it successfully detected one of car image in positive
images. Later on we publish this video to get detail fro yoour
learning.
- The problem is we don’t know the method Opencv did, what
features of images that opencv campare and train. That’s why, I
continue this similar project to be my final project assignment.

Adam Puspabhuana,S.Kom – 1st Semester, September 2014 Page 5


OpenCV and SVM developed by lecturer
- The lecturer gave us an example of getting feature and training
images using SVM technique.
-The method is getting pixel point of images in float format e.g.
detect Braille letter (3X2) pixel. Put that in 1X6 pixel(one image),
train using 6 images, it has (6X6) pixel.
-Then put the label of each images
-The application trained those images and tested by one image,
whether could recognize or not. The application gave a successfully
output, i.e could recognize label based on image tested.
-The problem is how we do train and recognize a lot of images with
large pixels. Then I try to figured out for my final assignment.

Adam Puspabhuana,S.Kom – 1st Semester, September 2014 Page 6


1st Application, Template Matching
- The original source from https://github.com/bytedeco/javacv
- JavaCV is such a library which is included all libraries needed.
- Add all libraries to your netbeans or eclipse.
- The application compare two images, first image is background
image (content any image), second image is the image that I crop
from background image, . E.g. Background image is lena.png, second
image(test image), matalena.png.
-The Output is make new image which is content image background
image and test image inside of it.
-- This is link of video that I uploaded on YouTube
-https://www.youtube.com/watch?v=hFAvjJoP_sk

Adam Puspabhuana,S.Kom – 1st Semester, September 2014 Page 7


2nd Application, Image Recognizer
- The original source from https://github.com/bytedeco/javacv
- JavaCV is such a library which is included all libraries needed.
- Add all libraries to your netbeans or eclipse.
- The application actually for recognizing faces, giving training image of faces
and testing it with one image sample of face.
- The result is getting label of image from training images that suitable with
testing image. Then I change the image, not face image but change it with car
images that I ever tried before for positive images. The result is successfully
predict label of image.
-The key is not in the function. But the function check with feature grayscale
and both training and testing image has same size of pic.
- This is link of video that I uploded on YouTube
-https://www.youtube.com/watch?v=h-e91GNg-rw

Adam Puspabhuana,S.Kom – 1st Semester, September 2014 Page 8


3rd Application, Face Recognition
- The original source from https://github.com/bytedeco/javacv
- JavaCV is such a library which is included all libraries needed.
- Add all libraries to your netbeans or eclipse.
- The application almost the same with 2nd application, the different
is given name for image itself and also check the list images which is
written in a text file. Both Training and testing images use text files.
- Size of pixel of images both training and testing has same size.
- The result is successfully recognize face images with grayscales of
faces.
- This is link of video that I uploaded on YouTube
-https://www.youtube.com/watch?v=h2P-stcAvPM

Adam Puspabhuana,S.Kom – 1st Semester, September 2014 Page 9


Closing
- JavaCV could be implemented cross platform such as android,
linux, mac and windows.
- JavaCV is a complete library which is developed and distributed
freely.
- From first experiment getting image detection until final
assignment, I learn a lot about computer vision features and
algorithm. Actually SVM techniques could be expanded such as BoW
(Bag of Words) or BoF(Bag of Features). Not only that there are two
more that I ever learn such as SURF and SIFT.
- Thank you for your attention and interest. I wish I could High score
for Computer Vision Subject of this semester.

Adam Puspabhuana,S.Kom – 1st Semester, September 2014 Page 10