Climatic and atmospheric phenomena – such as haze, fog and smoke – may lead to deterioration of quality and poor scenic clarity of outdoor images. In computer graphics, the authors can model these images as a linear combination of scene radiance, medium transmission and airlight. Several techniques have been proposed to remove the effects of haze from images using this model. The most effective approach for removing the haze effect from a single image is based on dark channel prior. Dark channel prior is based on statistical observation of outdoor images comprising some regions with dark intensity pixels. Here we propose a new l ₂-norm-based prior to generate a dark channel in order to remove the haze from a single-input image. The dark channel generated using this new prior is more robust and free from the block-effect. We also propose a statistical technique for airlight estimation of a given image. The proposed technique for modifying the dark channel prior and the airlight estimation are robust techniques as compared with approaches detailed in currently available literature. By combining this modified dark channel and estimated airlight, the haze can be directly removed and a more accurate haze-free image can be recovered from single-input hazy image.

Plastic surgery is considered as a challenging research issue in the field of face recognition. Nevertheless, it has yet to be studied from theoretical and experimental perspectives. In this study, the authors proposed a facial recognition system for recognising faces after plastic surgery, which fuses the scores of two feature-based and texture-based algorithms. The feature based algorithm is the image GIST global descriptor and the texture-based algorithm is the local binary pattern (LBP) of silence points. First, the local texture descriptor LBP was applied over a set of key points (silence points) in the face image rather than applying it over the entire face area. This proposed feature set is based on the assumption that only those LBP patterns with certain meaning, such as an edge or corner, will be useful for recognising faces that have undergone plastic surgery. The second set of features was extracted using a global descriptor, which is the GIST descriptor, to obtain a basic and a subordinate level description of the perceptual dimension. The performance of the proposed system surpassed the performance of a number of state-of-the-art face recognition after plastic surgery, with a maximum verification accuracy of more than 91%.

Recently, the development of automatic face annotation techniques in online social networks has become a promising research area for the purpose of management of the large numbers of photographs uploaded to social network platforms. In this study, the authors first construct the personalised pyramid database units for each member in the pyramid database access control module by effectively making use of various types of social network context to drastically reduce time expenditure and further boost the accuracy of face identification. Next, they train and optimise the personalised multiple-kernel learning (MKL) classifier unit for each member, which utilises the MKL algorithm to locally adapt to each member, resulting in the production of high-quality face identification results for the current owner in the MKL face recognition module. Experimental results demonstrate that their proposed face annotation approach provides a substantially higher level of efficacy and efficiency than other face annotation approaches for real-life personal photographs with pose variations.

Automated feature matching of nuclear detonations (NUDETs) enables three-dimensional point cloud reconstruction, and establishment a volume-based model to reduce uncertainty in estimating the yield of NUDETs solely from video. Establishing a volume-based model requires feature correspondences between wide viewpoints of 58°–110° that are larger than scale-invariant feature transform-based techniques can reliably match. The presented technique detects relative bright features in the NUDET known as ‘hotspots,’ and matches them across wide viewpoints using a spherical based object model. Results show that hotspots can be detected with a 71.95% hit rate and 86.03% precision. Hotspots are matched to films from different viewpoints with 76.6% correctness and a standard deviation of 16.4%. Hotspot descriptors are also matched in time sequence with 99.6% correctness and a standard deviation of 1.07%. The results demonstrate that a spherical model can serve as a viable descriptor model for matching across wide viewpoints when the object is known to be spherical. It also demonstrates an automated feature detection and matching combination that enables features to be matched from unsynchronised video across wide viewpoints of 58°–110° on spherical objects where state-of-the-art techniques are insufficient.

Appearance-based person re-identification is particularly difficult due to varying lighting conditions and pose variations across camera views. Taking inspiration from image retrieval, in which windowed searching over locations is proven to be more effective, the authors first perform dense local feature matching using graph cuts to properly deal with the pose variation problem. However, the re-identification problem suffers from far more overlap between feature distributions. In a re-identification problem, many samples cropped from surveillance videos are heavily contaminated by external factors or internal mechanical noises, making the images from the same pedestrian totally different. These overly difficult samples would significantly degenerate the training performance. To address this problem, a query-level loss function for ranking is proposed, benefiting from taking into account the training data every query set to decrease the punishment for those morbid samples. The authors further develop a coarse-to-fine iterative algorithm, where the update in each iteration is computed by solving a gradient-based optimisation and update iteration is to refine the training data by adjusting an ‘Expected rank’ parameter. The authors present experiments to demonstrate the performance gain of the proposed method over existing template matching and ranking models.

Coherent nature of crowd movement allows representing the crowd motion using sparse features. However, surveillance videos recorded at different periods of time are likely to have different crowd densities and motion characteristics. These varying scene properties necessitate use of different models for an effective representation of behaviour at different periods. In this study, a density aware approach is proposed to detect motion-based anomalies for scenes having varying crowd densities. In the training, the sparse features are modelled using separate hidden Markov models, each of which becomes an expert for specific scene characteristics. These models are then used for anomaly detection. The proposed method automatically adapts to the changing scene dynamics by switching to the most representative model at each frame. The authors demonstrate the effectiveness and real-time performance of the proposed method on real-life datasets as well as on simulated crowd videos that they generated and made publicly available to download.

The frontal-parallel assumption is made by many matching algorithms, but this assumption fails for slanted surfaces. This study proposes a matching algorithm intended to improve the matching results for slanted surfaces. First, a mathematical model is constructed to prove that slanted surfaces in the environment have corresponding slanted disparity surfaces in the disparity space image, and the model is to help find the proper plane parameters of slanted support windows, then improved cost aggregation and post-processing methods are proposed. The algorithm is tested using the Middlebury and Karlsruhe Institute of Technology and Toyota Technical Institute at Chicago (KITTI) benchmarks. The results demonstrate that the algorithm exhibits good performance and is efficient for slanted surfaces.

Automatic analysis of human behaviour in large collections of videos is rapidly gaining interest, even more so with the advent of file sharing sites such as YouTube. From one perspective, it can be observed that the size of feature vectors used for human action recognition from videos has been increasing enormously in the last five years, in the order of ∼100–500K. One possible reason might be the growing number of action classes/videos and hence the requirement of discriminating features (that usually end up to be higher-dimensional for larger databases). In this study, the authors review and investigate feature projection as a means to reduce the dimensions of the high-dimensional feature vectors and show their effectiveness in terms of performance. They hypothesise that dimensionality reduction techniques often unearth latent structures in the feature space and are effective in applications such as the fusion of high-dimensional features of different types; and action recognition in untrimmed videos. They conduct all the authors’ experiments using a Bag-of-Words framework for consistency and results are presented on large class benchmark databases such as the HMDB51 and UCF101 datasets.

Common VSM (vehicle speed measurement) methods lack either necessary precision and/or flexibility to be used conveniently in law enforcement applications. In this study, a new method is proposed for simultaneous camera calibration and VSM compensating for these problems. The contribution of this study is to give a solution to the degeneracy problem of the Zhang's algorithm while dealing with parallel planar patterns and using the calibration data to estimate vehicle's speed. In applications like estimating the vehicle's speed in straight roads in which the license plates are parallel to each other, the number of linearly independent equations is reduced to 2 and the method cannot be used. In this study, the degeneracy problem of the Zhang's algorithm in dealing with parallel patterns is solved. Using orthonormal properties of all rotation vectors, the number of linearly independent equations is increased from 2 to 5 and camera calibration and VSM are performed simultaneously. Results of implementing this algorithm on real images taken by law enforcement cameras show that the proposed method can estimate the vehicle's speed with a mean of errors less than 1.22 km/h and consequently is proper for law enforcement applications.

Recently crowded cities are having an increasing need for an advanced parking management system to help drivers to locate the vacant and available parking places in real time. In this study, the authors propose a novel multi agent parking lots management system based on vision approach to detect and localise the vacant parking places at a city level and to provide drivers with relevant information in real time. The elaboration of a multi agent approach for parking places modelling enable the cooperation between different agents in order to improve the results of the proposed system and to detect different cases of anomalies and abnormal situations that can be caused by the drivers, such as the cases where parked cars affect the state of more than one place or when a car blocks the access to parking lots.

Although second-order cone programming (SOCP) has been applied to optimise camera parameters in computer vision, it is occasionally been used to refine fisheye camera external parameters as well. This study presents a fisheye camera external parameter estimation based on SOCP in convex optimisation. The homography constraint between two spherical images are first exploited to derive an equation with respect to a given error threshold. Then, the fisheye camera external estimation is transformed into an SOCP optimisation problem through reformulating the parameter estimation equation. The SOCP method has been implemented in Matlab and the optimisation toolbox has been made publicly available. The fisheye camera external parameter optimisation method has been validated by some experiments with synthetic and real data. Comparison experiments between the proposed method and other methods in the literature are also carried out, and the results show that the SOCP method is better for the corrected images.

As a challenging problem, image enhancement plays an important role in computer vision applications and has been widely studied. As one of the most difficult issues of image enhancement, outdoor nighttime image enhancement suffers from noise amplification easily. To solve this problem, this study proposes a parameter-adaptive nighttime image enhancement method with multi-scale decomposition. The main contributions of this work are threefold. First, the authors find out that noises in different scales are various, and their method decomposes an input image into three high-frequency layers and a background layer accordingly. Second, the authors’ method enhances each high-frequency layer using adaptive parameters based on the characteristics of noises. Third, the proposed method maps the background layer to make it suitable to present details. Experiment results demonstrate that the proposed method can suppress noises as well as improve details effectively.

The authors present a sparsity-based algorithm, basic thresholding classifier (BTC), for classification applications which is capable of identifying test samples extremely rapidly and performing high classification accuracy. They introduce a sufficient identification condition (SIC) under which BTC can identify any test sample in the range space of a given dictionary. By using SIC, they develop a procedure which provides a guidance for the selection of threshold parameter. By exploiting rapid classification capability, they propose a fusion scheme in which individual BTC classifiers are combined to produce better classification results especially when very small number of features is used. Finally, they propose an efficient validation technique to reject invalid test samples. Numerical results in face identification domain show that BTC is a tempting alternative to sparsity-based classification algorithms such as greedy orthogonal matching pursuit and l ₁-minimisation.

Domain adaptation is used for machine learning tasks, when the distribution of the training (obtained from source domain) set differs from that of the testing (referred as target domain) set. In the work presented in this study, the problem of unsupervised domain adaptation is solved using a novel optimisation function to minimise the global and local discrepancies between the transformed source and the target domains. The dissimilarity in data distributions is the major contributor to the global discrepancy between the two domains. The authors propose two techniques to preserve the local structural information of source domain: (i) identify closest pair of instances in source domain and minimise the distances between these pairs of instances after transformation; (ii) preserve the naturally occurring clusters present in source domain during transformation. This cost function and constraints yield a non-linear optimisation problem, used to estimate the weight matrix. An iterative framework solves the optimisation problem, providing a sub-optimal solution. Next, using orthogonality constraint, an optimisation task is formulated in the Stiefel manifold. Performance analysis using real-world datasets show that the proposed methods perform better than a few recently published state-of-the-art methods.

Unmanned vehicles are autonomous robotic systems that are fully or partially controlled by an operator remotely from a station. In the last two decades, massive amount of advancements have been observed regarding unmanned vehicles for both military and civilian purposes. Today majority of these vehicles require human guidance even for basic missions, thus, minimising the human intervention on such systems is one of the emerging research topics. To serve this purpose, this study proposes a new trajectory tracking algorithm using Mirage pose estimation method. Mirage employs target pixel errors in two-dimensional image plane and analytically calculates the robot's pose in three-dimensional Euclidean space. Therefore, complex computations are not needed and undesirable Euclidean trajectories are avoided since the vehicle's pose is directly controlled. Both simulations and real experiments were performed to verify the effectiveness of the method. The results show that the proposed method is a feasible alternative for vision-based Euclidean trajectory tracking with high accuracy and low complexity.

Straight line matching is a fundamental task in many applications such as scene matching, stereo vision and sequence images analysis. As individual line segments cannot completely present the information of image and methods based on them are difficult to achieve high matching accuracy, the authors propose a straight line matching algorithm based on line pairs and feature points. Extracted lines are clustered into line feature sets according to their spatial proximity and geometric structures. The structural relationship between line pairs, which are selected from the line feature sets, is described by feature vector consisting of length ratio, angle and average gradient. Coarse matching of line segments is achieved by using feature vector as similarity measurement. To eliminate the mismatches in the matched straight lines, the constraints of feature points are employed. The experimental results demonstrate that the authors’ algorithm can successfully match image lines with high accuracy under various image transformations, including scale, illumination changes and viewpoint variations.