Stuttering is a prevalent speech disorder affecting millions worldwide. To provide an automatic and objective stuttering assessment tool, Stuttering Event Detection (SED) is under extensive investigation for advanced speech research and applications. Despite significant progress achieved by various machine learning and deep learning models, SED directly from speech signal is still challenging due to stuttering speech’s heterogeneous and overlapped nature. This paper presents a novel SED approach using multi-feature fusion and attention mechanisms. The model utilises multiple acoustic features extracted based on different pitch, time-domain, frequency domain, and automatic speech recognition feature to detect stuttering core behaviours more accurately and reliably. In addition, we exploit both spatial and temporal attention mechanisms as well as Bidirectional Long Short-Term Memory (BI-LSTM) modules to learn better representations to improve the SED performance. The experimental evaluation and analysis convincingly demonstrate that our proposed model surpasses the state-of-the-art models on two popular stuttering datasets, with 4% and 3% overall F1 scores, respectively. The superior results indicate the consistency of our proposed method, supported by both multi-feature and attention mechanisms in different stuttering events datasets.

There is a growing interest among the research community in the search for possible technology-driven strategies for the conservation of the much-needed, historically rich and culturally important, desert life. In this work, we investigate the use of one of the best available Deep Neural Networks, YOLO Version-5 (v5), to enable offline detection, identification and classification of three popular desert animals (i.e Camels, Oryxes, and Gazelles) in a Drone Imagery Dataset captured by the Dubai Desert Conservation Reserve (DDCR), United Arab Emirates. The dataset contains over 1200 images, which were partitioned into training, validation, and testing data sub-sets in a 8:1:1 ratio, respectively. We trained three multi-class models, animal classification models, based on YOLO v5 Small(S), Medium(M) and Large(L), representing increasingly deep and complex architectures, to simultaneously detect and label the 3 kinds of animals. Models' performance was compared on the basis of classification accuracy (F1-Measure), The multi-class detector models generated were also compared with the single animal detector models created using the same network architectures, to assess the trained network's robustness against detecting more than one class of object. YOLO v5 L achieved the highest multi-class average classification accuracy of 96.71 percent (95.39 - 98.98). In comparison with the single animal detector models, the multi-class models exhibited the ability to correctly detect the target objects even for cases where the objects are located close to each other. We show that the promising results achieved in this work provide a promising foundation for the development of real-time multiclass identification and classification applications utilizing UAV imagery, to aid in the conservation efforts of fauna, particularly in the urbanized modern-day deserts and semi-desert places, such as the DDCR. We provide comprehensive test results and an analysis of results to demonstrate the effectiveness of the proposed models.

The Ghaf is a drought-resilient tree native to some parts of Asia and the Indian Subcontinent, including the United Arab Emirates (UAE). To the UAE, the Ghaf is a national tree, and it is regarded as a symbol of stability and peace due to its historical and cultural importance. Due to increased urbanization and infrastructure development in the UAE, the Ghaf is currently considered an endangered tree, requiring protection. Utilization of modern-day aerial surveillance technologies in combination with Artificial Intelligence (AI) can particularly be useful in keeping count of the Ghaf trees in a particular area, as well as continuously monitoring unauthorized use to feed animals and to monitor their health status, thereby aiding in their preservation. In this paper, we utilize one of the best Convolutional Neural Networks (CNN), YOLO-V5, based model to effectively detect Ghaf trees in images taken by cameras onboard light-weight, Unmanned Aircraft Vehicles (UAV), i.e. drones, in some areas of the UAE. We utilize a dataset of over 3200 drone captured images partitioned into data-subsets to be used for training (60%), validation (20%), and testing (20%). Four versions of YOLO-V5 CNN architecture are trained using the training data subset. The validation data subset was used to fine tune the trained models in order to realize the best Ghaf tree detection accuracy. The trained models are finally evaluated on the reserved test data subset not utilized during training. The object detection results of the Ghaf tree detection models obtained by the use of four different sub-versions of YOLO-V5 are compared quantitatively and qualitatively. YOLO-V5x model produced the highest average detection accuracy of 81.1%. In addition, YOLO-V5x can detect and locate Ghaf trees of different sizes moreover in complex natural environments and in areas with sparse distributions of Ghaf trees. The promising results presented in this work offer fundamental grounds for AI-driven UAV applications to be used for monitoring the Ghaf tree in real-time, and thus aiding in its preservation.

The United Arab Emirates (UAE) values its relationship with the desert, considering it a crucial part of its heritage and culture. However, the desert faces environmental challenges due to the improper disposal of garbage by visitors and the dumping of waste, as some perceive the desert as an empty wasteland. The rise in tourism exacerbates the problem, as litter negatively impacts the desert's ecology, wildlife, and natural habitats. Traditional litter collection methods involving human patrols are inadequate for the vast desert terrain. Drones equipped with high-resolution cameras offer a potential solution by conducting aerial surveys quickly and efficiently. However, the manual review of drone footage to detect litter is time-consuming. This paper explores the use of deep neural network architectures, such as Faster R-CNN, SSD, and YOLO, to develop litter detection models. These models focus on distinguishing litter from other man-made objects. The training dataset consists of thousands of samples, and the models are evaluated based on their performance in detecting and locating litter in drone images captured at different altitudes and environmental conditions. The evaluation includes objective and subjective analyses. The research aims to alleviate the practical challenges of litter detection in the desert by automating the process through computer vision-based object detection methods.

Despite the success of state-of-the-art single image superresolution algorithms using deep convolutional neural networks in terms of both reconstruction accuracy and speed of execution, most proposed models rely on minimizing the mean square reconstruction error. More recently, inspired by transfer learning, Mean Square Error (MSE)-based content loss estimation has been replaced with loss calculated on feature maps of the pre-trained networks, e.g. VGG-net used for ImageNet classification. We demonstrate that this alternative approach is sub-optimal and adds false colour and mosaicking artefacts in the reconstructed images. In this paper, we present the first Convolutional Neural Network (CNN) capable of optimizing its parameters by minimizing the loss of in-network, self-features. To achieve this, we propose a new loss function for a light CNN architecture, which contains a mapping of residual blocks between the low and high resolution images. Our proposed method performs better than existing methods that apply perceptual loss by effectively suppressing false colour-effects. We show that the in-network features used for the determination of the loss function will give new insights for future research and applications when designing deep learning networks for other computer vision tasks such as demosaicing and denoising.

In this work, we examine the feasibility of applying Deep Convolutional Generative Adversarial Networks (DCGANs) with Single Shot Detector (SSD) as data-processing technique to handle with the challenge of pedestrian detection in the wild. Specifically, we attempted to use in-fill completion to generate random transformations of images with missing pixels to expand existing labelled datasets. In our work, GAN’s been trained intensively on low resolution images, in order to neutralize the challenges of the pedestrian detection in the wild, and considered humans, and few other classes for detection in smart cities. The object detector experiment performed by training GAN model along with SSD provided a substantial improvement in the results. This approach presents a very interesting overview in the current state of art on GAN networks for object detection. We used Canadian Institute for Advanced Research (CIFAR), Caltech, KITTI data set for training and testing the network under different resolutions and the experimental results with comparison been showed between DCGAN cascaded with SSD and SSD itself.

This paper presents a novel efficient two-region Selective encryption approach that exploits medical images statistical properties to adaptively segment Digital Imaging and Communications in Medicine (DICOM) images into regions using thresholding technique in the spatial domain. This approach uses adaptive pixel thresholding, in which thresholds for same DICOM modality, anatomy part and pixel intensities' range were extracted off-line. Then, the extracted thresholds were objectively and subjectively evaluated to select the most accurate threshold for the correspondent pixel intensities' range. In the on-line stage, DICOM images were segmented into a Region Of Interest (ROI) and a Region Of Background (ROB) based on their pixels intensities using the adopted thresholds. After that, ROI was encrypted using Advanced Encryption Standard (AES), while ROB was encrypted using XXTEA. The main goal of the proposed approach is to reduce the encryption processing time overhead in comparison with the Naïve approach; where all image pixels are encrypted using AES. The proposed approach aims to achieve a trade-off between processing time and a high level of security. The encryption time of the proposed approach can save up to 60% of the Naïve encryption time for DICOM images with small-medium ROI

Human action recognition from video sequences is a challenging topic in computer vision research. In recent years, many studies have explored the use of deep learning representations to consistently improve the analysis accuracy. Meanwhile, designing a fast and reliable framework is becoming increasingly important given the exponential growth of video data collected for many purposes (e.g. public security, entertainment, and early medical diagnosis etc.). In order to design a more efficient automatic human action annotation method, the sequential probability ratio test, one of the classical statistical sampling scheme, is adapted to solve a multi-classes hypothesis test problem in our work. With the proposed algorithm, the computational cost is reduced significantly without sacrificing the performance of the underlying system. The experimental results based on the UCF101 data set demonstrated the efficiency of the framework compared to the fixed sampling scheme.

In this paper, we present a novel framework for representation of images as a combination of multiple mid-level feature descriptor representation based group of visual words. The mid-level feature representation is computed on discriminative patches of the image to build a lexicon, the visual words of which are used to represent the shape within that image. The proposed image representation method has been applied to the application of vehicles make and model recognition. Each make, model class is represented as an over complete sub-lexicon of mid-level feature representation. The classification of vehicles is performed by comparing the visual words of probe image with the learned lexicon of training data using Euclidean distance. The proposed framework offers the advantage of accurate recognition in the presence of significant background clutter. The experiments have shown that the proposed representation successfully captures the fine-grained inter and intra-class discrimination to recognize the model and make of the vehicle without any strict requirement of precise region of interest segmentation. Another important contribution of the paper is a comprehensive dataset of cars depicting images collected in the wild.

Automatic baggage detection has become a subject of significant practical interest in recent years. In this paper, we propose an approach to baggage detection in CCTV video footage that uses color information to address some of the vital shortcomings of state-of-the-art algorithms. The proposed approach consists of typical steps used in baggage detection, namely, the estimation of moving direction of humans carrying baggage, construction of human-like temporal templates, and their alignment with the best matched view-specific exemplars. In addition, we utilize the color information to define the region that most likely belongs to a human torso in order to reduce the false positive detections. A key novel contribution is the person's viewing direction estimation using machine learning and shoulder shape related features. Further enhancement of baggage detection and segmentation is achieved by exploiting the CIELAB color space properties. The proposed system has been extensively tested for its effectiveness, at each stage of improvement, on PETS 2006 dataset and additional CCTV video footage captured to cover specific test scenarios. The experimental results suggest that the proposed algorithm is capable of superseding the functional performance of state-of-the-art baggage detection algorithms.

In this paper we propose an enhanced approach to automatic baggage detection in CCTV video footage that addresses some of the shortcomings of state-of-the-art algorithms and a novel approach to the subsequent recognition of their type. The proposed approach consists of standard steps used in baggage detection, namely, the estimation of moving direction of humans carrying baggage and the alignment of a temporal human-like template with the best matched view-specific exemplars. In addition, unique to the proposed approach is the connected baggage separation algorithm to achieve accurate baggage detection. A further key novel contribution is baggage type classification which is based on the determination of the position of the bag in relevance to the human body carrying it. The proposed system has been extensively tested for its effectiveness on PETS 2006 dataset and additional CCTV video footage captured to cover specific test scenarios. The experimental results suggest that the proposed algorithm is capable of superseding the functional performance of state-of-the art baggage detection algorithms and is able to achieve accurate baggage type classification.

Mobile imaging and digital storytelling currently support a growing practice of multimedia communication in the West. In this paper we describe a project which explores their benefit in the East, to support non-textual information sharing in an Indian village. Local audiovisual story creation and sharing activities were carried out in a one month trial, using 10 customized cameraphones and a digital library of stories represented on a village display. The findings show that the system was usable by a cross-section of the community and valued for its ability to express a mixture of development and community information in an accessible form. Lessons for the role of HCI in this context are also discussed.

Block based texture synthesis algorithms have shown better results than others as they help to preserve the global structure. Previous research has proposed several approaches in the pixel domain, but little effort has been taken in the synthesis of texture in a multiresolution domain. We propose a multiresolution framework in which coefficient-blocks of the spatio-frequency components of the input texture are efficiently stitched together to form the corresponding components of the output texture. We propose two algorithms to this effect. In the first, we use a constant block size throughout the algorithm. In the second, we adaptively split blocks so as to use the largest possible block size in order to preserve the global structure, while maintaining the mismatched error of the overlapped boundaries below a certain error tolerance. Special consideration is given to minimization of the computational cost, throughout the algorithm designs. We show that the adaptation of the multiresolution approach results in a fast, cost-effective, flexible texture synthesis algorithm that is capable of being used in modern, bandwidth-adaptive, real-time imaging applications. A collection of regular and stochastic test textures is used to prove the effectiveness of the proposed algorithm.

In this paper, we propose a hybrid scheme to implement an object driven, block based algorithm to achieve low bit-rate compression of stereo image pairs. The algorithm effectively combines the simplicity and adaptability of the existing block based stereo image compression techniques with an edge/contour based object extraction technique to determine appropriate compression strategy for various areas of the right image. Unlike the existing object-based coding such as MPEG-4 developed in the video compression community, the proposed scheme does not require any additional shape coding. Instead, the arbitrary shape is reconstructed by the matching object inside the left frame, which has been encoded by standard JPEG algorithm and hence made available at the decoding end for those shapes in right frames. Yet the shape reconstruction for right objects incurs no distortion due to the unique correlation between left and right frames inside stereo image pairs and the nature of the proposed hybrid scheme. Extensive experiments carried out support that significant improvements of up to 20% in compression ratios are achieved by the proposed algorithm in comparison with the existing block-based technique, while the reconstructed image quality is maintained at a competitive level in terms of both PSNR values and visual inspections.