Implementation of high performance feature extraction method using oriented fast and rotated brief algorithm

IJRET: International Journal of Research in Engineering and Technology eISSN: 2319-1163 | pISSN: 2321-7308
_______________________________________________________________________________________
Volume: 04 Issue: 02 | Feb-2015, Available @ http://www.ijret.org 394
IMPLEMENTATION OF HIGH PERFORMANCE FEATURE
EXTRACTION METHOD USING ORIENTED FAST AND ROTATED
BRIEF ALGORITHM
Prashant Aglave1
, Vijaykumar.S. Kolkure2
1
M.E (Electronics, Appeared), Department Of Electronics Engineering, Bharatratna Indira Gandhi College of
Engineering, Affiliated to Solapur University, Solapur, Maharashtra, India.
2
Assistant Professor, Department Of Electronics Engineering, Bharatratna Indira Gandhi College of Engineering,
Affiliated to Solapur University, Solapur, Maharashtra, India.
Abstract
Feature-based image matching is an important characteristic in many computer based applications such as object recognition,
3D stereo reconstruction, structure-from-motion and images stitching. These applications require a lot real-time performance.
Feature based algorithms are well-suited for such operations. Different algorithms are used for image processing like Scale-
invariant feature transform (SIFT), Speeded up Robust Features (SURF), Oriented FAST and Rotated BRIEF (ORB). ORB is one
of the fast binary descriptor which is relying on BRIEF, where the BRIEF is rotation invariant and resistant to noise. This paper
gives the advantages of rotation invariance and scale invariance of ORB algorithm for object detection technique. Query based
object detection method is explained in this paper for object detection with efficient computation time. Different experimental
results prove the scale invariance and rotation in variance of ORB in query based object detection method.
Keywords: ORB, BRIEF, SIFT, SURF
-------------------------------------------------------------------***-------------------------------------------------------------------
1. INTRODUCTION
Now days many industrial applications require object
recognition and tracking capabilities. Many feature-based
algorithms are well-suited for such operations. Recent
industrial applications needs object recognition and tracking
capabilities. Object recognition technology has matured to a
point at which exciting applications are becoming possible.
Oriented FAST and Rotated BRIEF (ORB) is a very fast
binary descriptor which is relying on BRIEF. Where BRIEF
is rotation invariant and resistant to noise. ORB is a two
order of magnitude faster than Scale-invariant feature
transform (SIFT), it can be verified through experiments
while performing as well in many situations.
ORB uses the known FAST keypoint detector and the
BRIEF descriptor, that why we call it as ORB (Oriented
FAST and Rotated BRIEF). Both Fast keypoint detector and
BRIEF descriptor are attractive because of they have better
performance and low cost. The recognize method for object
recognition is Scale invariant feature transform (SIFT),
which is popular for its invariance to scaling, rotation and
illumination, is computationally difficult due to its weighty
workload required in local feature extraction and matching
operation. Thus computer vision kind of applications
demands high performance and low complexity solution and
ORB provides better solution to it.
2. GENERALIZED BLOCK DIAGRAM
ORB is basically a fusion of FAST keypoint detector and
BRIEF descriptor with many modifications to enhance the
performance. First it uses FAST to find keypoints, and then
apply Harris corner measure to find top N points among
them. It also use pyramid to produce multiscale-features.
ORB is rotation invariant and resistant to noise. Another
advantage of ORB is its very low memory requirement. Its
descriptor provides comparable precision/recall results with
SURF and SIFT.
Following figure explains the methodology to be used in
proposed system. The proposed system can be divided into
following steps:
1) Conversion of image to text files using MATLAB.
2) Applying ORB algorithm to the image.
3) Conversion of text files to image in MATLAB.
4) Displaying the output image.
Fig 2.1 Generalized Block Diagram

_______________________________________________________________________________________
For object detection we require two input images: Trained
image and query image. ORB algorithm is then applied on
each image to extract the interest points in each image. Fig
2.1shows the generalized block diagram of object detection
using ORB. After extracting the features of both images,
they are matched to detect object in the query image.
3. FLOW OF PROPOSED WORK
Proposed work can be divided into two main parts first one
the Simulation flow and second one is its implementations.
In this proposed work we will take the input images both
simple image and rotated image, then we will find the corner
points in both simple and rotated images using ORB
detector. After that we describe the features in binary form
using ORB descriptor and find all the matches in both
simple and rotated images. Out of all the matches sort out
the good matches, then we will draw the match and show
the result. After that repeat this procedure with different
rotation angles in second image.
Fig 3.1Simulation Flow of proposed work
3.1 Feature Matching Algorithm
Feature matching algorithm includes following steps:
1. Take both simple and rotated input images-
In this step feature matching simple and rotated image is
taken.
2. Calculate the features of both the images-
ORB uses FAST detector to detect features from images.
FAST keypoint detector takes an input and calculates the
corner points of the image. Cornet points have the property
that, there exists a set of `n' contiguous pixels in the circle
(of 16 pixels) which are all brighter than Intensity of pixel
point added with threshold, or all darker than intensity of
pixel subtracted by threshold. Function for comparing
intensity values is given as:
3. Find out the binary descriptors for the features in the
images. [12]
(a) In a 31x31 area round an interest point.
(b) Randomly create 256 9x9 pairs patches, call them
Ai, Bi.
(c) For each pair, if, then set the corresponding bit to 1, else
0.
(d) The resulting 256 bit vector is the descriptor for the
patch.
4. Match both the descriptors vectors-
The Hamming distance between two strings of equal length
is the number of positions at which the corresponding
symbols are different. In other way, it measures the
minimum number of substitutions required to change one
string into the other, or the minimum number of errors that
could have transformed one string into the other.
Eg. The Hamming distance between 1011101 and 1001001
is 2.
5. Find out the common points in the image-
If the images are same then find out the common points.
6. Find the good matches.
Select the strongest keypoints depending upon the distance.
Find the homography between the two kepoints, which will
give good matches.
7. Plot the matched good keypoints in the images.
Finally plot the good keypoints in the images.
4. OBJECT DETECTION
Query based object detection method is explained in figure
4.1. In this method, ORB algorithm is applied to an image.
Features described by this image are used to detect whether
the object in this image is present in the other image or not.
This image is called as ‘Trained Image’. The image in which
the object is to be detected is known as ‘Query Image’.
After calculating features of both the images using ORB,
this method matches the common features in the two images
and draws the box around the object if features are matched.

_______________________________________________________________________________________
Fig 4.1 Query Based Object Detection Flow Chart
Above flow chart shows query based object detection.
Following steps are used to detect an object.
1. To detect the object in an image, we need minimum two
images:
(a) Image of the object to be detected (trained image).
(b) Image in which the object is to be detected (Query
image).
2. Detect Feature Points using the ORB detector.
3. Extract Feature Descriptors by passing the keypoints
(features) as input to the ORB descriptor.
4. Find putative point matches using appropriate matcher.
5. Locate the object in the query image Using Putative
Matches.
6. Detect Another Object.
5. RESULTS
Fig 5.1 Object detection using ORB
The first image is the trained image and second image is the
query image. Here we have to find the trained image in
query image. Fig c. shows the object detection using ORB
algorithm. We also done the object detection with ‘case.png’
using different angle rotation and the result are shown
below.
Table 5.1 shows rotation invariance in the query based
object detection method with image ‘case.png’.
Computation time required for the object detection is
calculated.
Table 5.1: Rotation Invariance Table for ‘case’ image
SR.No Image Rotation
(Degrees)
Computation
Time(Seconds)
1 Case.png 45 1.259
2 Case.png 50 1.446
3 Case.png 75 1.514
4 Case.png 110 1.372
In table 5.1 change in the computation time is because of the
change in size of the query image due to rotation at different
angles. Table 5.2 shows scale invariance results in the query
based object detection method with image ‘case.png’. The
algorithm is applied on different sizes of trained image and
computation time is calculated for each size of the image.

_______________________________________________________________________________________
Table 5.2: Scale invariance Table for ‘case’ image
SR.No Image Image
Size
Computation
Time(Seconds)
1 Case.png 300x300 1.177
2 Case.png 200x200 1.278
3 Case.png 100x100 1.663
4 Case.png 50x50 1.421
Query based object detection method detects trained image
object of different sizes with efficient computation time.
Change in the computation time is because of the resizing of
the image. Table 5.1 shows rotation invariance in the query
based object detection method with image ‘case.png’.
Computation time required for the object detection is
calculated.
6. CONCLUSION
This proposed system will help to extract the features in a
given image by using ORB irrespective of the rotation of the
image. Features extracted using this algorithm will be
further used for classification and detection of the object
also for the annotation of the image. ORB implementation
consists of fast detector which has fast and accurate
orientation component to fast and BRIEF descriptor. ORB is
better than SIFT and SURF to find adjacent-neighbor
matching over large databases of images. Proposed query
based object detection method is tested for different
rotations and sizes of the trained image. Experimental
results show that this method detects the object for any
rotation angle and size of the trained image. Thus, proposed
query based object detection method which uses ORB for
feature extraction solves the problem of rotation invariance
and scale in variance to maximum extent in object detection.
REFERENCES
[1] Ethan Rublee, Vincent Rabaud, Kurt Konolige Gary
Bradski, “ORB: an efficient alternative to SIFT or
SURF”.
[2] OndrejMiksik, “Evaluation of Local Detectors and
Descriptors for Fast Feature Matching".
[3] Michael B. Holte, Cuong Tran, Mohan M. Trivedi,
Thomas B. Moeslund,”Human Pose Estimation and
Activity Recognition From Multi-View Videos:
Comparative Explorations of Recent Developments".
[4] Herbert Bay, Tinne Tuytelaars, and Luc Van
Gool,”SURF: Speeded Up Robust Features".
[5] Yan Ke, Rahul Sukthankar, “PCA-SIFT: A More
Distinctive Representation for Local Image
Descriptors", International Journal of Conceptions on
Electronics and Communication Engineering,Vol. 1,
Issue. 1, Dec 2013; ISSN: 2357 2809.
[6] ViniVidyadharan, and SubuSurendran, “Automatic
Image Registration using SIFT-NCC”, Special Issue
of International Journal of Computer Applications
(0975 – 8887) , pp.29-32, June 2012.
[7] D. Lowe. “Distinctive Image Features from Scale-
Invariant Keypoints”, Accepted for publication in the
International Journal of Computer Vision, pp. 1-28,
2004.
[8] Edward Rosten and Tom Drummond, “Machine
learning for high-speed corner detection",
International Conference on Computer Vision,2008.
[9] Edward Rosten and Tom Drummond, “Fusing points
and lines for high performance tracking",European
Conference on Computer Vision,2006.
[10] Dilip K. Prasad, “Survey of The Problem of Object
detection In Real Images",International Journal of
Image Processing(IJIP),Volume(6):Issue(6):2012.
[11] Mikolajczyk, K., Schmid, “ An affine invariant
interest point detector.ECCV.(2002) 128.
[12] S.Winder and M. Brown, “Learning Local Image
Descriptors,” in Proc.CVPR’07, 2007
BIOGRAPHIES
Prashant Aglave1
has received his B.E.
degree from, University of Pune in 2008.
He is currently pursuing M.E in
Electronics Engineering from B.I.G.C.E,
Solapur University.
Vijaykumar S. Kolkure 2
has completed M.E. (Electronics)
from W.C. Sangli, Maharashtra, India. He has 10 years of
teaching experience. Currently he is working as Assistant
Professor at Bharatratna Indira Gandhi College of
Engineering, Solapur, Maharashtra, India. His area of
interest is Image Processing, Video Processing.

Implementation of high performance feature extraction method using oriented fast and rotated brief algorithm

Recommended

More Related Content

What's hot (20)

Viewers also liked (17)

Similar to Implementation of high performance feature extraction method using oriented fast and rotated brief algorithm (20)

More from eSAT Journals (20)

Recently uploaded (20)

Implementation of high performance feature extraction method using oriented fast and rotated brief algorithm