An Image-Based Approach to Detection of Fake Coins

ABSTRACT:

We propose a new approach to detect fake coins using their images in this paper. A coin image is represented in the dissimilarity space, which is a vector space constructed by comparing the image with a set of prototypes. Each dimension measures the dissimilarity between the image under consideration and a prototype. In order to obtain the dissimilarity between two coin images, the local keypoints on each image are detected and described. Based on the characteristics of the coin, the matched keypoints between the two images can be identified in an efficient manner. A postprocessing procedure is further proposed to remove mismatched keypoints. Due to the limited number of fake coins in real life, one-class learning is conducted for fake coin detection, so only genuine coins are needed to train the classifier. Extensive experiments have been carried out to evaluate the proposed approach on different datasets. The impressive results have demonstrated its validity and effectiveness.

PROJECT OUTPUT VIDEO:

EXISTING SYSTEM:

• In recent years, a lot of illegal counterfeiting rings manufacture and sell fake coins, which have caused great loss and damage to the society. Thus it is imperative to be able to detect fake coins. This is also an important concern in the field of numismatics. Forensic experts may be employed to examine the suspected coins, yet it is unrealistic considering the large quantities of coins that have to be examined. Therefore, an automatic fake coin detection system is highly desired.

• In recent years, the local keypoint detectors and descriptors have been widely employed to describe an image. The common pipeline is to first detect some keypoints in the image using detectors like Difference-of-Gaussian (DOG), Harris-Laplace, Harris-Affine, Hessian-Laplace or Hessian-Affine. Subsequently, the image region within a certain radius around the keypoint is described using descriptors such as Scale Invariant Feature Transform (SIFT), PCA-SIFT, gradient location and orientation histogram (GLOH) and shape context (SC).

• The local keypoint detectors and descriptors are distinguished by their great discriminative power and robustness to image distortions such as illumination, resolution and viewpoint transformations. Besides, the number of keypoints varies across images according to their characteristics. However, representing an image in terms of a set of keypoints cannot fit in the vector-based machine learning tools. To deal with this problem, the Bag-Of-Visual- Words (BOVW) model proposed by Sivic and Zisserman may be employed.

• Analogous to the Bag-Of-Words (BOW) model in the text domain, a visual word vocabulary is built through clustering the local descriptors obtained from a training set. Hence an arbitrary descriptor can be represented by its nearest visual word in the vocabulary. Consequently, an image is represented by a vector, the dimension being equal to the size of the vocabulary.

DISADVANTAGES OF EXISTING SYSTEM:

• In spite of the popularity of the BOVW model, the discriminative power of the local descriptors is substantially diminished due to the involved clustering process.

PROPOSED SYSTEM:

• In this study, we also employ the local descriptors to generate image representations because of their superb discriminative power. However, instead of resorting to the BOVW model for generation of vectorial representations, we represent a coin image in the dissimilarity space

• The dissimilarity space is constructed based on a very basic operation, viz. comparison. It benefits from mimicking human perceptions well. When presented a suspected coin, we humans tend to compare it with its genuine counterparts to see whether they are different or not. Likewise, to construct the dissimilarity space, each coin image to be detected will be compared with some prototype coins that are genuine and selected beforehand, based on which the image can be represented as a vector. The dimension of the vector is determined by the number of prototypes and each dimension measures the dissimilarity between the coin image under consideration and one of the prototypes.

• The matched keypoints between the two images can be identified e_ciently by taking into account the characteristics of the coins. Furthermore, we propose a postprocessing procedure to remove mismatched keypoints. The dissimilarity between the two images is then derived from the number of matched keypoints. In this study, we compare two di_erent prototype selection methods: random selection and clustering-based selection. The random selection approach, as its name suggests, selects a set of prototypes randomly, while the K-medoids clustering [36] is employed for prototype selection in the clustering-based selection approach.

• Considering the fact that the number of fake coins is usually very limited in the real world, one-class learning is conducted for fake coin detection. We employ one-class SVM, which adapts the SVM methodology proposed by Vapnik to the one-class scenario, so only genuine coins are needed to train the classifier.

ADVANTAGES OF PROPOSED SYSTEM:

• Since no clustering is applied to the local descriptors like in the BOVW model, we are able to achieve the best of both worlds: the great discriminative power of the local keypoint descriptors and the availability of the machine learning tools.

• The superiority of using dissimilarity space over the BOVW model in generating image vectorial representations is evident from the experiments presented.

• The proposed approach is evaluated extensively on four di_erent datasets, containing coins that are of di_erent denominations and from di_erent countries.

• Besides, the impact of the parameters involved in the proposed approach is thoroughly investigated. We also compare it with the fake coin detection method in the literature. The promising results have confirmed the potential of the proposed approach.

MODULES:

1. Image Acquisition

2. Preprocessing

3. Feature Extraction

4. Dissimilarity Measure
5. Classification

MODULE DESCRIPTION:

1. Image Acquisition:

• Image acquisition in image processing can be broadly defined as the action of retrieving an image from some source, usually a hardware-based source, so it can be passed through whatever processes need to occur afterward.

• Performing image acquisition in image processing is always the first step in the workflow sequence because, without an image, no processing is possible. The image that is acquired is completely unprocessed and is the result of whatever hardware was used to generate it, which can be very important in some fields to have a consistent baseline from which to work.

2.Preprocessing:

• In preprocessing stage, coin detection is explored. Coin extraction is done by using binary image masking.

• First, the color image is converted into gray image then binary mask was generated. Finally coin region was extracted from background using binary masking.

3. Feature Extraction:

• In feature extraction, we used two features for classify of fake coin that features are DOG and SIFT.

• we first detect keypoints on the coin image based on the DOG detector. Afterwards, the SIFT descriptor is chosen for keypoint description. The combination of DOG detector and SIFT descriptor has been shown to outperform other detector & descriptor combinations in many applications.

4. Dissimilarity Measure:

• Similarly real coin key points also extracted to dissimilarity measure. Given two coin images, their dissimilarity can be derived from the number of matched keypoints.

• Generally speaking, the smaller the number of matched keypoints, themore dissimilar the two coin images are. So the descriptor associated with the keypoint on one coin is compared with the descriptors of all the keypoints on the other coin in terms of Euclidean distance, from which its closest and second-closest neighbors can be identified.

• Based on the distance ratio of closest to second-closest neighbors, a decision with respect to whether the keypoint under consideration and its closest neighbor keypoint on the other coin are matched or not can be made by comparing the ratio with a threshold.

• Finally key points matching done by geometric verification.

5. Classification:

• The classification process is done over the extracted features. The main novelty here is the adoption of SVM.

• SVM classifier is applied over the features and the classification is done.

• Finally get the result is given coin is fake or real by using SVM.

SYSTEM REQUIREMENTS:

HARDWARE REQUIREMENTS:

• System          : Pentium Dual Core.

• Hard Disk      : 120 GB.

• Monitor          : 15’’ LED

• Input Devices : Keyboard, Mouse

• Ram                : 1GB.

SOFTWARE REQUIREMENTS:

• Operating system : Windows 7.

• Coding Language : MATLAB

• Tool                       : MATLAB R2013A

REFERENCE:

Li Liu, Yue Lu, Senior Member, IEEE, and Ching Y. Suen, Fellow, IEEE, “An Image-Based Approach to Detection of Fake Coins”, IEEE TRANSACTIONS ON INFORMATION FORENSICS AND SECURITY, 2017.