Application countries flags recognition in image processing
Abstract
In this paper a system for world country flags recognition and detection is presented, which are based on the extract image feature, then search in database for matching results. The proposed system start with reading any input image, then demonstrate whether it was country flag or not and recognize to which country it belongs. The procedure of identifying the country name consist several stages such reading input image, image processing and feature extraction and matching. In the paper, we introduce our methodology, testing the system, analysis the outcomes and how much is the system strong to detect different images as possible. After testing the results shows very high success in flag recognition.
Keywords: object detection, object recognition, matching process, canny edge detector, signature , skeleton, Hough transform, Harris corner detector.
Introduction
In this paper, we present a robust recognition system for countries flags based on object detection and object recognition technologies. The system establishment done on MTALB program which provide user friendly interface for huge image processing tools and functions. Which will manipulate input image to detect and recognize if there is a flag in the image or not and demonstrate which country the flag belong.in order to achieve the requirements of the system we use multiple image features that can help for matching and description process like edge detection, color, signature, compactness and so on.
Object detection and recognition technologies in image processing and computer vision fields are very essential and play an important role to detect objects and recognize them in image as possible in a robust manner whatever the image situation is. In the recent years, Object detection and recognition are very trendy and many researches and applications in different area have been and take advantage from like face detection, disease detection, text recognition and many others.
Object detection technology responsible to detect semantic of objects of a specified class in an image.
Object recognition technology responsible to distinguish and identify objects in an image or video sequence.
Related work
In the past years, there has been much designing and research work on object recognition filed like fingerprint recognition [15] [16].Face recognition [9] proposed methodology is assessed with the face recognition task, [14] presented a fully automatic face recognition system based on probabilistic decision-based neural networks. place recognition [13] presented robust place recognition by exploit visual context to perform .people detector [12] presented the component-based person detection system for static images that can detect frontal, rear, slightly rotated and partially occluded people in cluttered scenes without having any a priori knowledge concerning the image. And other system based on object recognition[11] presented a novel discriminative model for effective and efficient recognition and semantic segmentation of objects in images.
Most of these systems use the same workflow as the one proposed in this work. Starting with image preprocessing technique, use segmentation after that use feature extraction like in [10][17][18][19][20][21]they use keypoints, Hough transform, color histograms, image signature, Canny edge detector and detect edges respectively, then the last stage match the feature from input image with those stored ones, finally testing to measure how the robust of the system.
Applications of image detection and object recognition
Image detection and object recognition have many applications in wide areas such as robotics, medical, transportation areas, security systems and so on. One of interesting application is face recognition that used for security purpose in companies, banks, military. security applications to recognize bombs or explosives. also, object detection can be used for people counting, to analyses a store performance or airport due to the cowed. There is also application in medical such as, detect breast cancer and broken bones. in industrial processes the object detection is used also to identify products.
Challenges of object detection and object recognition
1- Illumination: changing light intensity in image make considerable difference in it and how the object can be seen.
2- Occlusion: is anything that distort vision. the problem from occlusion comes when object parts can be difficult to reconstructed again due to occlusion.
3- Scale
4- View point variation of object.
5- Deformation: indicates any changes in the shape or size of an object in image.
6- Intra-Class Variation: its refer to the randomness of object itself. As more random found recognition process become easy.
7- Background clutter: it means noise which degrade the image quality.
Object Features methods
We use multiple features algorithm to gain useful data about objects in flag image.
��� Canny edge detection
Canny edge detection it is an important algorithm that aims to extract useful structural information about object boundaries and reduce the amount of image data as well as remove noise [2]. The developer of canny algorithm is John F. Canny in 1986.canny provide the three main criteria of edge detection which are: -good detection with low error rate, good localization in which the detected point localized on the center of the edge, and low spurious response to immune detector to produce multiple maxima [3]. since canny guarantee these three criteria its became one of most reliable method to detect edges. canny algorithm works in multi-stage process. first it applies a Gaussian filter to smooth all the images to remove noise. then compute gradient using gradient operator (Roberts, Sobel, Prewitt,). After that suppress non-maxima pixels in the edges. then apply double threshold and then trace each edge segment to suppressing edges that are weak and not connected to strong edges.
Figure 1 shows an example of canny edge detector applied on Afghanistan flag.
Figure 1: canny edge detector on Afghanistan flag.
��� Harris corner detection
A corner (also in general its known as interest point) in digital image is an intersect between two edges. An interest point simply means any point in the image for which the signal changes two-dimensionally [5]. the interest point has defined position in the image and can be strongly detected. So when corner detect, we first detect interest points and then fetch which in of these points are corner points. Harris corner detector is a robust well known corner detector sue it is in variance to: scale, image noise, rotation and illumination variation.it works with local auto-correlation function that measures local changes of the signals in different directions [6].
Figure (2) display Afghanistan flag after applying harries corner detection algorithm, it highlights strongest 50 point in the image.
Figure (2): harries corner detection on Afghanistan flag image.
��� Hough Transform
Hough transform [7]is a robust technique to detect regular curves such as lines, circles or ellipse in images. to use Hough transform to detected straight lines in image we use the line segment formula
r=x cos (��)+ x sin(��)
Where r is the distance from the origin to the line a long vector perpendicular to the line and �� is the angel between the vector and the x-axis. The Hough transform produce matrix who's the dimensions corresponds to r and ��. Then find the peak values that represent possible strongest lines in image and the endpoint of line segment that identical to these peaks.
Figure 2 shows Hough transform after apply Hough function on Afghanistan flag binary image.
Figure (2): Hough transform on Afghanistan flag binary image.
Figure 3 display original image with the longest lines segments.
Figure (3): original image with the longest line segments.
��� Eulre number
Eulre number is a scalar that represents the total number of objects in image minus total number of holes in those objects. It works with binary images. and its very powerful feature to include since its invariance to size, translation and rotation.
��� Skeletons (skeletonizing):
significant way to representing the structural shape of plane region is to reduce it to a graph. The medial axis transformation (MAT) is one way to define the skeleton. The MAT is for each point in region find the closest point in border of region and if there more than one such point we said that point belong to MAT. Skeleton is a good feature to extract because each shape has its own one for example the skeleton for circle is a point [8].
Figure (4) display Italy flag before and after applying skeleton algorithm on it .
(A) (B)
Figure (4): A- Italy flag B- skeleton for Italy
��� signature:
represent boundary in one dimensional function and can be generated in several ways like plotting the distance from center to the boundary as function of angles. The basic idea in signature is to convert boundary representation from two to one dimension function to make it easy to describe. Signature is normalized to being translation and scale invariant and can besimpli1ed by quantizing into a signature histogram, which is rotationally invariant. Figure (5) display Italy flag after finding signature for it.
Figure (5): signature for Italy
System Approach
In this paper, MATLAB program is used since it provides a comprehensive toolbox from functions, algorithms and applications for image processing and computer vision. In our system, we create a database contains 224 images flags for different countries. Then we extract multiple features from each image. these features are color, canny edge detector, Harris corner detection, Hough transform, skeletons, signature, signature. These features are applied on all images. After that we store all feature vectors and do matching process ��� ��� which based on comparing vectors of all features with feature vectors that belongs to sample images to test how strong our system is. The number of samples we use is . then a classification process is applied and the system made the decision.
Experiment result
Conclusion
Image processing techniques are very important in enhancing images quality
[2] Canny, John. "A computational approach to edge detection." IEEE Transactions on pattern analysis and machine intelligence 6 (1986): 679-698.���
[3] Bao, Paul, Lei Zhang, and Xiaolin Wu. "Canny edge detection enhancement by scale multiplication." IEEE transactions on pattern analysis and machine intelligence 27.9 (2005): 1485-1490.���
[4] Patil, Jayamala K., and Raj Kumar. "Advances in image processing for detection of plant diseases." Journal of Advanced Bioinformatics Applications and Research 2.2 (2011): 135-141.���
[5] Schmid, Cordelia, Roger Mohr, and Christian Bauckhage. "Evaluation of interest point detectors." International Journal of computer vision 37.2 (2000): 151-172.���
[6] Derpanis, Konstantinos G. "The harris corner detector." York University (2004).���
[7] Duda, Richard O., and Peter E. Hart. "Use of the Hough transformation to detect lines and curves in pictures." Communications of the ACM 15.1 (1972): 11-15.���
[8] Papasaika-Hanusch, Haris. "Digital image processing using matlab."Institute of Geodesy and Photogrammetry, ETH Zurich 63 (1967).
[9] Ahonen, Timo, Abdenour Hadid, and Matti Pietikainen. "Face description with local binary patterns: Application to face recognition." IEEE transactions on pattern analysis and machine intelligence 28.12 (2006): 2037-2041.
[10] Lowe, David G. "Distinctive image features from scale-invariant keypoints."International journal of computer vision 60.2 (2004): 91-110.
[11] Shotton, Jamie, et al. "Textonboost for image understanding: Multi-class object recognition and segmentation by jointly modeling texture, layout, and context." International Journal of Computer Vision 81.1 (2009): 2-23.
[12] Mohan, Anuj, Constantine Papageorgiou, and Tomaso Poggio. "Example-based object detection in images by components." IEEE transactions on pattern analysis and machine intelligence 23.4 (2001): 349-361.
[13] Torralba, Antonio, et al. "Context-based vision system for place and object recognition." Computer Vision, 2003. Proceedings. Ninth IEEE International Conference on. IEEE, 2003.
[14] S.-H. Lin, S.-Y. Kung, and L.-J. Lin, Face recognition/detection by probabilistic decision-based neural network, IEEE Trans. Neural Networks 8, 1997, 114���132.
[15] Knapp, Alan G. "Fingerprint sensing device and recognition system having predetermined electrode activation." U.S. Patent No. 5,325,442. 28 Jun. 1994.
[16] Jea, Tsai-Yang, and Venu Govindaraju. "A minutia-based partial fingerprint recognition system." Pattern Recognition 38.10 (2005): 1672-1684.
[17] Ballard, Dana H. "Generalizing the Hough transform to detect arbitrary shapes." Pattern recognition 13.2 (1981): 111-122.
[18] Zhang, Zhengyou, et al. "A robust technique for matching two uncalibrated images through the recovery of the unknown epipolar geometry." Artificial intelligence 78.1 (1995): 87-119.
[19] Overgaard, Niels. "On a modification to the harris corner detector."Symposium Svenska S��llskapet f��r Bildanalys, Stockholm. 2003.
[20] Hou, Xiaodi, Jonathan Harel, and Christof Koch. "Image signature: Highlighting sparse salient regions." IEEE transactions on pattern analysis and machine intelligence 34.1 (2012): 194-201.
[21] Bao, Paul, Lei Zhang, and Xiaolin Wu. "Canny edge detection enhancement by scale multiplication." IEEE transactions on pattern analysis and machine intelligence 27.9 (2005): 1485-1490.
[21] Canny, John. "A computational approach to edge detection." IEEE Transactions on pattern analysis and machine intelligence 6 (1986): 679-698.