3. Electronic Theses and Dissertations (ETDs) - All submissions
Permanent URI for this communityhttps://wiredspace.wits.ac.za/handle/10539/45
Browse
7 results
Search Results
Item Fake image detection using machine learning(2018-03-23) Jama, Ahmed, MohamoudThe rapid growth of digital image processing technologies and editing software has given rise to large amounts of tampered images circulating in our daily lives. This undermines credibility and trustworthiness of digital images and also creates false beliefs in many real world situations. Hence it is generating a great demand for automatic forgery detection algorithms in order to determine the authenticity of a candidate image in a timely fashion. Many techniques and tools have been implemented to detect such type of forgery with the real image but because of increasing editing software every day, the problem is not solved yet. In this thesis, we will first introduce some information about digital photographs, their formats and how they are compressed, we will also introduce some important tools that were implemented to detect image forgery and how well these tools can detect faked or tampered images. In this work we wish to test whether a convolutional neural network can be trained to efficiently detect image forgeries. The CNN will be trained first using two classes (real and fake), and then using the same CNN architecture with three classes (real, copy-paste and spliced images). The network will be trained and tested using the standard Casia 2:0 dataset. Then we will compare the convolutional neural network that we trained with some of the implemented tools that classifies images into real and fake. We observe a very good performance after the training. That means the trained CNN network is able to recognize the tampered images at a maximum success rate of 85:63%. So the use of this application can be used as a false proof technique in digital authentication. It will also greatly reduce spreading of fake images through websites and through social media.Item Joint decoding of parallel power line communication and visible light communication systems(2018) Onwuatuelo, Daniel ObinnaMany indoor applications operate at narrow band (3kHz148.5kHz) speed and for such applications, power line communication (PLC) and visible light communication (VLC) networks can be naturally connected and adapted to complement each other in order to gain more overall system performance in terms of bit error rate (BER) and computational complexity. In this research,the joint decoding of parallel PLC and VLC systems is proposed and its BER performance is compared to that of the PLCa nd the VLC systems. The joint decoding is applied either at the inner (Viterbi) or at the outer (Reed-Solomon) decoder. The proposed system is adopted according to the PLC G3 physical layer specification but direct current optical orthogonal frequency division multiplexing OFDM (DCO-OFDM) is used in the VLC system to ensure that only positive (unipolar) signals are transmitted. A realistic VLC channel model is adopted in this research by considering the VLC channel as an additive white Gaussian noise (AWGN) channel affected by attenuation in terms of angle of orientation between the source and the receiver and effective surface area of the receiver. Furthermore, the PLC channel is modeled as an AWGN channel with background and impulsive noise generated using Middleton Class Anoisedistributionmodel. Itisshownthroughsimulationresultsandanalysisthatthe proposed joint decoded system outperforms the PLC and the VLC systems in terms of BERperformancedependingonthedistanceofseparationbetweenthesourceandthe receiver. Key words: Power line communication (PLC), Visible light communication (VLC), Bit error rate (BER), Joint decoding, Orthogonal frequency division multiplexing (OFDM), DCopticalOFDM(DCO-OFDM),AdditivewhiteGaussiannoise(AWGN).Item Deformable part model with CNN features for facial landmark detection under occlusion(2018) Brink, HannoDetecting and localizing facial regions in images is a fundamental building block of many applications in the field of affective computing and human-computer interaction. This allows systems to do a variety of higher level analysis such as facial expression recognition. Facial expression recognition is based on the effective extraction of relevant facial features. Many techniques have been proposed to deal with the robust extraction of these features under a wide variety of poses and occlusion conditions. These techniques include Deformable Part Models (DPMs), and more recently deep Convolutional Neural Networks (CNNs). Recently, hybrid models based on DPMs and CNNs have been proposed considering the generalization properties of CNNs and DPMs. In this work we propose a combined system, using CNNs as features for a DPM with a focus on dealing with occlusion. We also propose a method of face localization allowing occluded regions to be detected and explicitly ignored during the detection step.Item Conservation photography: evaluating the impact of photographic images in conveying conservation related messages(2018) Soma Owen, PrelenaSpecies extinction and environmental threats continue to be a concern for conservation scientists. A limitation to reaching conservation targets is the challenge in communicating scientific findings to the general public. Conservation photography is increasingly acknowledged by both scientists and conservationists as an effective tool in communicating biodiversity losses and environmental concerns. This study explored the impact of photographic images in delivering conservation messaging on a purposely selected sample group and tested if demographics played a role in image interpretation. The study found that photos of charismatic animals did not rank high in delivering effective conservation messages. Respondents chose images that contained graphic content, with “shock value” as having the strongest conservation messages. This contradicted general expectations. The study also found that 50 % of the images used in the study showed statistical significance in the manner in which they were interpreted by Black and White respondents, suggesting that demographics played a role in image interpretation. Both these findings have important implications for conservation communication.Item Using customised image processing for noise reduction to extract data from early 20th century African newspapers(2017) Usher, SarahThe images from the African articles dataset presented challenges to the Optical Character Recognition (OCR) tool. Despite successful binerisation in the Image Processing step of the pipeline, noise remained in the foreground of the images. This noise caused the OCR tool to misinterpret the text from the images and thus needed removal from the foreground. The technique involved the application of the Maximally Stable Extremal Region (MSER) algorithm, borrowed from Scene-Text Detection, and supervised machine learning classifiers. The algorithm creates regions from the foreground elements. Regions are classifiable into noise and characters based on the characteristics of their shapes. Classifiers were trained to recognise noise and characters. The technique is useful for a researcher wanting to process and analyse the large dataset. They could semi-automate the foreground noise-removal process using this technique. This would allow for better quality OCR output, for use in the Text Analysis step of the pipeline. Better OCR quality means less compromises would be required at the Text Analysis step. These concessions can lead to false results when searching noisy text. Fewer compromises means simpler, less error-prone analysis and more trustworthy results. The technique was tested against specifically selected images from the dataset which exhibited noise. It involved a number of steps. Training regions were selected and manually classified. After training and running many classifiers, the highest performing classifier was selected. The classifier categorised regions from all images. New images were created by removing noise regions from the original images. To discover whether an improvement in the OCR output was achieved, a text comparison was conducted. OCR text was generated from both the original and processed images. The two outputs of each image were compared for similarity against the test text. The test text was a manually created version of the expected OCR output per image. The similarity test for both original and processed images produced a score. A change in the similarity score indicated whether the technique had successfully removed noise or not. The test results showed that blotches in the foreground could be removed, and OCR output improved. Bleed-through and page fold noise was not removable. For images affected by noise blotches, this technique can be applied and hence less concessions will be needed when processing the text generated from those images.Item Comparison of object and pixel-based classifications for land-use and land cover mapping in the mountainous Mokhotlong District of Lesotho using high spatial resolution imagery(2016) Gegana, MphoThe thematic classification of land use and land cover (LULC) from remotely sensed imagery data is one of the most common research branches of applied remote sensing sciences. The performances of the pixel-based image analysis (PBIA) and object-based image analysis (OBIA) Support Vector Machine (SVM) learning algorithms were subjected to comparative assessment using WorldView-2 and SPOT-6 multispectral images of the Mokhotlong District in Lesotho covering approximately an area of 100 km2. For this purpose, four LULC classification models were developed using the combination of SVM –based image analysis approach (i.e. OBIA and/or PBIA) on high resolution images (WorldView-2 and/or SPOT-6) and the results were subjected to comparisons with one another. Of the four LULC models, the OBIA and WorldView-2 model (overall accuracy 93.2%) was found to be more appropriate and reliable for remote sensing application purposes in this environment. The OBIA-WorldView-2 LULC model was subjected to spatial overlay analysis with DEM derived topographic variables in order to evaluate the relationship between the spatial distribution of LULC types and topography, particularly for topographically-controlled patterns. It was discovered that although that there are traces of the relationship between the LULC types distributions and topography, it was significantly convoluted due to both natural and anthropogenic forces such that the topographic-induced patterns for most of the LULC types had been substantial disrupted.Item Application of pattern recognition to projective 3D image processing problems.(2014-03-12) Danaila, Mariana LianaThis dissertation presents the development and performance of a few algorithms used for automated scene matching. The objective is to recognise and predict the location of a template (reference image) inside a degraded scene image (sensed image). A set of perspective, projective optical images of relatively well defined man-made objects located in areas of varying background is used as database. Perturbations to the grey levels of the image cause artefacts that easily destroy the unique match location and generate false fixes. Therefore, suitable enhancement and noise removal techniques are applied first. Several different types of features are investigated to decide upon those that are best suited to describe the original content of the scene. Statistical features, such as invariant moments are chosen for one of the algorithms, Multibcmd Ima^e using Moments (MBIMOM). The second one, Spatial Multiband Image (SMBI) algorithm, uses the spatial correlation of the pixels within a neighbourhood as initial descriptive features. Each algorithm uses either Principal Components transform or Maximum Noise Fraction transform for dimensionality and noise reduction. A normalised correlation coefficient of 1.00 was achieved by the SMBI algorithm. The final design of the algorithms is a trade-off between speed and accuracy.