A new perceptual dissimilarity measure for image retrieval and clustering
- Authors: Shojanazeri, Hamid
- Date: 2018
- Type: Text , Thesis , PhD
- Full Text:
- Description: Image retrieval and clustering are two important tools for analysing and organising images. Dissimilarity measure is central to both image retrieval and clustering. The performance of image retrieval and clustering algorithms depends on the effectiveness of the dissimilarity measure. ‘Minkowski’ distance, or more specifically, ‘Euclidean’ distance, is the most widely used dissimilarity measure in image retrieval and clustering. Euclidean distance depends only on the geometric position of two data instances in the feature space and completely ignores the data distribution. However, data distribution has an effect on human perception. The argument that two data instances in a dense area are more perceptually dissimilar than the same two instances in a sparser area, is proposed by psychologists. Based on this idea, a dissimilarity measure called, ‘mp’, has been proposed to address Euclidean distance’s limitation of ignoring the data distribution. Here, mp relies on data distribution to calculate the dissimilarity between two instances. As prescribed in mp, higher data mass between two data instances implies higher dissimilarity, and vice versa. mp relies only on data distribution and completely ignores the geometric distance in its calculations. In the aggregation of dissimilarities between two instances over all the dimensions in feature space, both Euclidean distance and mp give same priority to all the dimensions. This may result in a situation that the final dissimilarity between two data instances is determined by a few dimensions of feature vectors with relatively much higher values. As a result, the dissimilarity derived may not align well with human perception. The need to address the limitations of Minkowski distance measures, along with the importance of a dissimilarity measure that considers both geometric distance and the perceptual effect of data distribution in measuring dissimilarity between images motivated this thesis. It studies the performance of mp for image retrieval. It investigates a new dissimilarity measure that combines both Euclidean distance and data distribution. In addition to these, it studies the performance of such a dissimilarity measure for image retrieval and clustering. Our performance study of mp for image retrieval shows that relying only on data distribution to measure the dissimilarity results in some situations, where the mp’s measurement is contrary to human perception. This thesis introduces a new dissimilarity measure called, perceptual dissimilarity measure (PDM). PDM considers the perceptual effect of data distribution in combination with Euclidean distance. PDM has two variants, PDM1 and PDM2. PDM1 focuses on improving mp by weighting it using Euclidean distance in situations where mp may not retrieve accurate results. PDM2 considers the effect of data distribution on the perceived dissimilarity measured by Euclidean distance. PDM2 proposes a weighting system for Euclidean distance using a logarithmic transform of data mass. The proposed PDM variants have been used as alternatives to Euclidean distance and mp to improve the accuracy in image retrieval. Our results show that PDM2 has consistently performed the best, compared to Euclidean distance, mp and PDM1. PDM1’s performance was not consistent, although it has performed better than mp in all the experiments, but it could not outperform Euclidean distance in some cases. Following the promising results of PDM2 in image retrieval, we have studied its performance for image clustering. k-means is the most widely used clustering algorithm in scientific and industrial applications. k-medoids is the closest clustering algorithm to k-means. Unlike k-means which works only with Euclidean distance, k-medoids gives the option to choose the arbitrary dissimilarity measure. We have used Euclidean distance, mp and PDM2 as the dissimilarity measure in k-medoids and compared the results with k-means. Our clustering results show that PDM2 has perfromed overally the best. This confirms our retrieval results and identifies PDM2 as a suitable dissimilarity measure for image retrieval and clustering.
- Description: Doctor of Philosophy
- Authors: Shojanazeri, Hamid
- Date: 2018
- Type: Text , Thesis , PhD
- Full Text:
- Description: Image retrieval and clustering are two important tools for analysing and organising images. Dissimilarity measure is central to both image retrieval and clustering. The performance of image retrieval and clustering algorithms depends on the effectiveness of the dissimilarity measure. ‘Minkowski’ distance, or more specifically, ‘Euclidean’ distance, is the most widely used dissimilarity measure in image retrieval and clustering. Euclidean distance depends only on the geometric position of two data instances in the feature space and completely ignores the data distribution. However, data distribution has an effect on human perception. The argument that two data instances in a dense area are more perceptually dissimilar than the same two instances in a sparser area, is proposed by psychologists. Based on this idea, a dissimilarity measure called, ‘mp’, has been proposed to address Euclidean distance’s limitation of ignoring the data distribution. Here, mp relies on data distribution to calculate the dissimilarity between two instances. As prescribed in mp, higher data mass between two data instances implies higher dissimilarity, and vice versa. mp relies only on data distribution and completely ignores the geometric distance in its calculations. In the aggregation of dissimilarities between two instances over all the dimensions in feature space, both Euclidean distance and mp give same priority to all the dimensions. This may result in a situation that the final dissimilarity between two data instances is determined by a few dimensions of feature vectors with relatively much higher values. As a result, the dissimilarity derived may not align well with human perception. The need to address the limitations of Minkowski distance measures, along with the importance of a dissimilarity measure that considers both geometric distance and the perceptual effect of data distribution in measuring dissimilarity between images motivated this thesis. It studies the performance of mp for image retrieval. It investigates a new dissimilarity measure that combines both Euclidean distance and data distribution. In addition to these, it studies the performance of such a dissimilarity measure for image retrieval and clustering. Our performance study of mp for image retrieval shows that relying only on data distribution to measure the dissimilarity results in some situations, where the mp’s measurement is contrary to human perception. This thesis introduces a new dissimilarity measure called, perceptual dissimilarity measure (PDM). PDM considers the perceptual effect of data distribution in combination with Euclidean distance. PDM has two variants, PDM1 and PDM2. PDM1 focuses on improving mp by weighting it using Euclidean distance in situations where mp may not retrieve accurate results. PDM2 considers the effect of data distribution on the perceived dissimilarity measured by Euclidean distance. PDM2 proposes a weighting system for Euclidean distance using a logarithmic transform of data mass. The proposed PDM variants have been used as alternatives to Euclidean distance and mp to improve the accuracy in image retrieval. Our results show that PDM2 has consistently performed the best, compared to Euclidean distance, mp and PDM1. PDM1’s performance was not consistent, although it has performed better than mp in all the experiments, but it could not outperform Euclidean distance in some cases. Following the promising results of PDM2 in image retrieval, we have studied its performance for image clustering. k-means is the most widely used clustering algorithm in scientific and industrial applications. k-medoids is the closest clustering algorithm to k-means. Unlike k-means which works only with Euclidean distance, k-medoids gives the option to choose the arbitrary dissimilarity measure. We have used Euclidean distance, mp and PDM2 as the dissimilarity measure in k-medoids and compared the results with k-means. Our clustering results show that PDM2 has perfromed overally the best. This confirms our retrieval results and identifies PDM2 as a suitable dissimilarity measure for image retrieval and clustering.
- Description: Doctor of Philosophy
Imbalanced data classification and its application in cyber security
- Authors: Moniruzzaman, Md
- Date: 2020
- Type: Text , Thesis , PhD
- Full Text:
- Description: Cyber security, also known as information technology security or simply as information security, aims to protect government organizations, companies and individuals by defending their computers, servers, electronic systems, networks, and data from malicious attacks. With the advancement of client-side on the fly web content generation techniques, it becomes easier for attackers to modify the content of a website dynamically and gain access to valuable information. The impact of cybercrime to the global economy is now more than ever, and it is growing day by day. Among various types of cybercrimes, financial attacks are widely spread and the financial sector is among most targeted. Both corporations and individuals are losing a huge amount of money each year. The majority portion of financial attacks is carried out by banking malware and web-based attacks. The end users are not always skilled enough to differentiate between injected content and actual contents of a webpage. Designing a real-time security system for ensuring a safe browsing experience is a challenging task. Some of the existing solutions are designed for client side and all the users have to install it in their system, which is very difficult to implement. In addition, various platforms and tools are used by organizations and individuals, therefore, different solutions are needed to be designed. The existing server-side solution often focuses on sanitizing and filtering the inputs. It will fail to detect obfuscated and hidden scripts. This is a realtime security system and any significant delay will hamper user experience. Therefore, finding the most optimized and efficient solution is very important. To ensure an easy installation and integration capabilities of any solution with the existing system is also a critical factor to consider. If the solution is efficient but difficult to integrate, then it may not be a feasible solution for practical use. Unsupervised and supervised data classification techniques have been widely applied to design algorithms for solving cyber security problems. The performance of these algorithms varies depending on types of cyber security problems and size of datasets. To date, existing algorithms do not achieve high accuracy in detecting malware activities. Datasets in cyber security and, especially those from financial sectors, are predominantly imbalanced datasets as the number of malware activities is significantly less than the number of normal activities. This means that classifiers for imbalanced datasets can be used to develop supervised data classification algorithms to detect malware activities. Development of classifiers for imbalanced data sets has been subject of research over the last decade. Most of these classifiers are based on oversampling and undersampling techniques and are not efficient in many situations as such techniques are applied globally. In this thesis, we develop two new algorithms for solving supervised data classification problems in imbalanced datasets and then apply them to solve malware detection problems. The first algorithm is designed using the piecewise linear classifiers by formulating this problem as an optimization problem and by applying the penalty function method. More specifically, we add more penalty to the objective function for misclassified points from minority classes. The second method is based on the combination of the supervised and unsupervised (clustering) algorithms. Such an approach allows one to identify areas in the input space where minority classes are located and to apply local oversampling or undersampling. This approach leads to the design of more efficient and accurate classifiers. The proposed algorithms are tested using real-world datasets. Results clearly demonstrate superiority of newly introduced algorithms. Then we apply these algorithms to design classifiers to detect malwares.
- Description: Doctor of Philosophy
- Authors: Moniruzzaman, Md
- Date: 2020
- Type: Text , Thesis , PhD
- Full Text:
- Description: Cyber security, also known as information technology security or simply as information security, aims to protect government organizations, companies and individuals by defending their computers, servers, electronic systems, networks, and data from malicious attacks. With the advancement of client-side on the fly web content generation techniques, it becomes easier for attackers to modify the content of a website dynamically and gain access to valuable information. The impact of cybercrime to the global economy is now more than ever, and it is growing day by day. Among various types of cybercrimes, financial attacks are widely spread and the financial sector is among most targeted. Both corporations and individuals are losing a huge amount of money each year. The majority portion of financial attacks is carried out by banking malware and web-based attacks. The end users are not always skilled enough to differentiate between injected content and actual contents of a webpage. Designing a real-time security system for ensuring a safe browsing experience is a challenging task. Some of the existing solutions are designed for client side and all the users have to install it in their system, which is very difficult to implement. In addition, various platforms and tools are used by organizations and individuals, therefore, different solutions are needed to be designed. The existing server-side solution often focuses on sanitizing and filtering the inputs. It will fail to detect obfuscated and hidden scripts. This is a realtime security system and any significant delay will hamper user experience. Therefore, finding the most optimized and efficient solution is very important. To ensure an easy installation and integration capabilities of any solution with the existing system is also a critical factor to consider. If the solution is efficient but difficult to integrate, then it may not be a feasible solution for practical use. Unsupervised and supervised data classification techniques have been widely applied to design algorithms for solving cyber security problems. The performance of these algorithms varies depending on types of cyber security problems and size of datasets. To date, existing algorithms do not achieve high accuracy in detecting malware activities. Datasets in cyber security and, especially those from financial sectors, are predominantly imbalanced datasets as the number of malware activities is significantly less than the number of normal activities. This means that classifiers for imbalanced datasets can be used to develop supervised data classification algorithms to detect malware activities. Development of classifiers for imbalanced data sets has been subject of research over the last decade. Most of these classifiers are based on oversampling and undersampling techniques and are not efficient in many situations as such techniques are applied globally. In this thesis, we develop two new algorithms for solving supervised data classification problems in imbalanced datasets and then apply them to solve malware detection problems. The first algorithm is designed using the piecewise linear classifiers by formulating this problem as an optimization problem and by applying the penalty function method. More specifically, we add more penalty to the objective function for misclassified points from minority classes. The second method is based on the combination of the supervised and unsupervised (clustering) algorithms. Such an approach allows one to identify areas in the input space where minority classes are located and to apply local oversampling or undersampling. This approach leads to the design of more efficient and accurate classifiers. The proposed algorithms are tested using real-world datasets. Results clearly demonstrate superiority of newly introduced algorithms. Then we apply these algorithms to design classifiers to detect malwares.
- Description: Doctor of Philosophy
- «
- ‹
- 1
- ›
- »