Sharif Digital Repository

Providing security for digital information in a communication channel is a challenge in the recent years. This is due to the strength of high-speed processing resources that can threaten the security and correctness of communicating information. Using steganography method, the secret information can be hidden in an innocent media that does not attract the attention of third parties. This research discusses about steganography in images. The main goal of this research is presenting a universal steganography scheme with high steganography capacity. Using this scheme the steganographer can embed every large secret data in images securely. The reasons for selecting the subject of “Improving... 

Recently many works has been focused on Automatic Security Systems. Car type and color recognition systems are needed in vehicle based access control systems in buildings, outdoor sites and even housing estates. Another usage of car type and color recognition is when a certain model of car is being investigated in highways and streets. In this thesis two independent systems are designed; one for car type recognition and another for car color recognition. They both require to know the license plate location, so we proposed a license plate detection system. In the license plate detection system, first the plate candidates are extracted from image by gradient and morphological operations, then... 

Nowadays, digital images are easy to manipulate and edit due to availability of powerful image processing and editing software. It is possible to add or remove important features to the image without leaving any obvious traces of tampering. Therefore, proving the authenticity and integrity of digital media becomes increasingly important. In this study, we focus on detection of a special type of digital forgery, the copy-move attack, in which a part of the image is copied and pasted somewhere else in the image with the intent to cover an important image features or to add some fake feature. In this way we propose some novel methods that work in spatial domain, Discrete Cosine domain and... 

To overcome challenges such as object appearance changes and environment illumination variations in tracking methods, online algorithms are suggested to be used instead of offline ones. Online algorithms update the model by the information acquired in the last processed frame. The main challenge of using online algorithms is the accumulation of small errors after several steps of updating of the model (drift) which disturbs the model and causes tracking failure. Using the object information in the first frame in each update can be considered as a solution. The proposed online semi-supervised boosting algorithms can overcome the drift problem at the expense of decreasing their capabilities in... 

In this project, using one regular color camera, a video is captured of an actor’s face. Using machine vision and without any kind of markers on the actor’s face, this video is processed to extract the locations of some desired points on the face of the actor. Then these points are mapped to corresponding points on a 3D graphical model of a face, so that a realistic animation of facial movements is achieved. The extracted points, including points on the eyebrows, eyes and lips are the most important ones for the purpose of facial animation. To extract these points, in each region several methods are implemented and studied to find the best method, finally for extracting points on the... 

Motion planning in an autonomous vehicle is responsible for providing smooth, safe and efficient actions. Besides reducing the risk of collision with static and moving obstacles, the ability to make suitable decisionsencountering sudden changes in environment is very important. Many solutions for dealing this problem have been offered, one of which is, Artificial Potential Fields (APF). APF is a simple and computationally low cost method which keeps the robot away from the obstacles in the environment. However, this approach suffers from trapping in local minima’s of potential function and then fails to produce a plan. Furthermore,Oscillation in presence of obstacles or in narrow passages... 

Steganalysis is the art of detecting hidden message in a cover such as an image. All steganalysis methods either are designed for a specific steganographer or are blind. Since for a stego image, the steganography method is not available, this is important to detect a stego image without any knowledge about the steganography method using which the secret image was embedded. Therefore in order to dominate all steganography method, if a given image is a stego one, we need to use voting over several steganalysis methods applied on the stego image to improve the accuracy of detection. But this approach needs a long time to process that is not practical for most steganalysis applications. We know... 

The traditional machine learning methods assume that the training data and the test data are drawn from the same distribution (or drawn from the same domain). In practice, in many computer vision applications, this assumption may not hold. Unfortunately, the performance of these methods degrades on dataset drawn from a different domain. Domain adaptation attempts to minimize this degradation caused by distribution mismatch between the training and test data. Domain adaptation tries to adapt a model trainded from one domain to another domain. We focus on supervised domain adaptation method in which limited labeled data is available from the target domain. We propose a new domain adaptation... 

Tracking is one of the old and still not thoroughly solved problems in machine vision. Its importance lies on its many applications. These applications vary from security surveillance to examining the motion pattern of atomic particles. There is not a tracker which has acceptable results in all situations, yet. A tracker faces many difficulties such as change in illumination and occlusion. In past, tracking was done by using filters or optical flows. By use of the advances in machine learning and pattern recognition, many models have been proposed to accomplish tracking by using these new learning methods. In this dissertation, we proposed a new tracking method which utilizes sparse... 

Nowadays the number of digital images has largely increased because of progress in internet technology. Management of this volume of data needs an efficient system for browsing, categorizing, and searching the images. The goal of this research is to design a system for automatic annotation of unobserved images for better search in image data bases. Automatic image annotation is a multi-label classification problem with many labels which suggests some words for describing the content of an image. Designing AIA systems faces chanllenges like semantic gap between low level image features and high level human expressions (tags), incompelete tags and imbalance images per tags in the datasets.... 

Melanoma is a malignant skin cancer which is caused by cancerous growth of melanocytes. If not treated at its early development stages, melanoma is the deadliest form of cancer. The best way to cure melanoma is to treat it in its earliest stage of development. Since a melanoma leasion is similar to benign moles (regaring its shape and appearance) at its early stages of development, it is often mistaken for moles and left untreated. Therefore, automatic melanoma detection can increase the survival rate of patients by detecting melanoma in its early stages. In this thesis, a new method for automatic diagnosis of melanoma using segmented dermoscopic images is provided. Almost all related... 

Lung cancer is one of the most common types of cancers, and its early diagnosis can save many lives. Due to the high number of computed tomography (CT) images used to detect lung cancer, it is difficult to accurately and rapidly diagnose this disease. Doing so requires high expertise by radiologists. Therefore the demand for computer aided diagnosis systems in this area has been increased. The core of all lung cancer detection systems is the distinction between cancer and non-cancerous tissues. The main objective of this study is to present a new method based on 3D convolutional neural networks (CNN) that can perform false positives reduction operations while providing high sensitivity. In... 

In this project we tried to to solve the problem of clustring images of a large image database. Considering that there is no prior information regarding domain of the images, we will review unsupervised clustring methods. For this, we will discuss about image description vector and similarity measures. At last, our contribution will be about finding the optimum number of clusters with the least of user invervention. Results of runnig our method on a databse with 1000 images is reported and compared to a similar method named CLUE. Our result shows considerable improvements when user feedback taken to account.
 

3D face modeling is currently a popular area in Computer Graphics and Computer Vision. Many techniques have been introduced for this purpose, such as using one or more cameras, 3D scanners, and many other systems of sophisticated hardware with related software. But the main goal is to find a good balance between visual reality and the cost of the system. In this thesis, reconstruction of a 3D human face from a pair of orthogonal views, front face and side face is studied. Unlike many other systems, facial feature points are obtained automatically from two photographs with the help of an Active shape model algorithm for the frontal face and an edge detection algorithm for side view of the... 

Content Based Image Retrieval (CBIR) is a research area in computer vision. This area comprises of two main steps, low level feature extraction such as color, texture and shape extraction and also similarity measures for comparison of images. The challenge in this system is the existence semantic gap between the low level visual features and the high level image semantics. The aim of research in this field is to reduce this semantic gap. In this study the images are divided into regions using Meanshift method, for color segmentation and then moments of each region as color feature are calculated. Also for extracting texture the images are divided into regions using Jseg method, and then... 

The advent of Internet and advancement of computer technologies have enabled convenient and fast exchange of multimedia so the illegal reproduction and modification of digital media has become increasingly serious. Hence, how to protect the intellectual property rights of digital multimedia is an imperative issue. Digital watermarking is one of the solutions to prevent unauthorized use of images. Traditional digital watermarking techniques embed a watermark such as logo, trademark, or copyright information into a host image so that it is not perceptible. These techniques, depending on the amount of embedded data, will distort the content of host image which results in quality degradation of... 

Digital watermarking is an efficacious technique to protect the copyright and ownership of digital information. Traditional image watermarking algorithms embed a logo in the image that could reduce its visual quality. A new approach in watermarking called zero watermarking doesn’t need to embed a logo in the image. In this algorithm we find a feature from the main image and combine it with a logo to obtain a key. This key is securely kept by a trusted authority. In this thesis we show that we can increase the robustness of digital zero watermarking by a new counter detection method in comparison to Canny Edge detection and morphological dilatation that is mostly used by related works.... 

Detection degradation caused by distribution discrepancy between the training and testing domains is a common problem in object detection systems. The difference between training and testing domains’ distribution mainly happenes because of the different ways of collecting and gathering data. For instance, datasets which have images with different illumination, view point, resolution, background and are obtained by different acquisition systems, have variance in distribution. The solution toward improving the detection rate of the classifier trained on training (source) domain when it is applied on testing (target) domain is to use Domain Adaptation (DA) techniques. One of important branches... 

It is not possible to compute a linear classifier to classify real world images, which are the focus of this thesis. Therefore, the space of such images is considered as a complex. In such cases, kernel trick in which data samples are implicitly mapped to a higher dimension space, leads to a more accurate classifier in such spaces. In kernel learning methods, the best kernel is trained for the classification problem in hand. Multiple Kernel Learning is a framework which uses weighted sum of multiple kernels. This framework achieves good accuracy in image classification since it allows describing images via various features. In the image input space which is composed of different extracted... 

Human detection in still images is one of the hardest problems in object detection area. There are several challenges like articulation, pose variation, variant clothing, none uniform illumination, cluttered background and occlusion which make this problem more sophisticated than any other object detection problem. The general solution for this kind of problems is based on supervised learning that contains two main parts: 1-extracting proper features, 2- using proper classifier. The main focus of this thesis is on the first part, extracting proper features, which should be robust to mentioned challenges. Based on the level of extraction we can divide the features to four groups: 1-low-level,...