Research in texture feature approximation is still in the embryonic stage because of difficulties in developing a sound theoretical model to express the unique pattern in the intensity-variation of pixels in the neighbourhood of the pixel-of-interest so that it can sufficiently discriminate different textures. Local texture descriptors are widely used in image segmentation as they comprise pixel-wise features. The Weber local descriptor (WLD) with differential excitation and gradient orientation components, inspired by Weber's Law, has been leveraged in the state-of-the-art iterative contraction and merging (ICM) image segmentation technique. However, WLD has inherent drawbacks in the formulation of the components that limit its discriminatory capability. This paper introduces a novel texture descriptor by directly modelling the distribution of intensity-variation in the parametric space of the Weibull distribution using its shape and scale parameters. A unified ‘joint scale’ texture property is introduced, which can discriminate textures better than the individual parameters while keeping the length of the descriptor shorter. Additionally, the accuracy of WLD's gradient orientation component is improved by using an extended Sobel operator and expressing gradients in$[-\pi /2,\pi /2)$range. When incorporated in ICM, the proposed texture descriptor has consistently outperformed WLD and a recent enhancement with radial mean WLD (RM-WLD) on three benchmark datasets. It has also outperformed two other texture segmentation techniques and their deep learning based improvements.

To exploit high temporal correlations in video frames of the same scene, the current frame is predicted from the already-encoded reference frames using block-based motion estimation and compensation techniques. While this approach can efficiently exploit the translation motion of the moving objects, it is susceptible to other types of affine motion and object occlusion/deocclusion. Recently, deep learning has been used to model the high-level structure of human pose in specific actions from short videos and then generate virtual frames in future time by predicting the pose using a generative adversarial network (GAN). Therefore, modelling the high-level structure of human pose is able to exploit semantic correlation by predicting human actions and determining its trajectory. Video surveillance applications will benefit as stored “big” surveillance data can be compressed by estimating human pose trajectories and generating future frames through semantic correlation. This paper explores a new way of video coding by modelling human pose from the already-encoded frames and using the generated frame at the current time as an additional forward-referencing frame. It is expected that the proposed approach can overcome the limitations of the traditional backward-referencing frames by predicting the blocks containing the moving objects with lower residuals. Our experimental results show that the proposed approach can achieve on average up to 2.83 dB PSNR gain and 25.93% bitrate savings for high motion video sequences compared to standard video coding.

Bidirectional mapping-based generalized zero-shot learning (GZSL) methods rely on the quality of synthesized features to recognize seen and unseen data. Therefore, learning a joint distribution of seen-unseen classes and preserving the distinction between seen-unseen classes is crucial for GZSL methods. However, existing methods only learn the underlying distribution of seen data, although unseen class semantics are available in the GZSL problem setting. Most methods neglect retaining seen-unseen classes distinction and use the learned distribution to recognize seen and unseen data. Consequently, they do not perform well. In this work, we utilize the available unseen class semantics alongside seen class semantics and learn joint distribution through a strong visual-semantic coupling. We propose a bidirectional mapping coupled generative adversarial network (BMCoGAN) by extending the concept of the coupled generative adversarial network into a bidirectional mapping model. We further integrate a Wasserstein generative adversarial optimization to supervise the joint distribution learning. We design a loss optimization for retaining distinctive information of seen-unseen classes in the synthesized features and reducing bias towards seen classes, which pushes synthesized seen features towards real seen features and pulls synthesized unseen features away from real seen features. We evaluate BMCoGAN on benchmark datasets and demonstrate its superior performance against contemporary methods.

Domain adaptation aims to transfer knowledge from a domain with adequate labeled samples to a domain with scarce labeled samples. Prior research has introduced various open set domain adaptation settings in the literature to extend the applications of domain adaptation methods in real-world scenarios. This paper focuses on the type of open set domain adaptation setting where the target domain has both private (‘unknown classes’) label space and the shared (‘known classes’) label space. However, the source domain only has the ‘known classes’ label space. Prevalent distribution-matching domain adaptation methods are inadequate in such a setting that demands adaptation from a smaller source domain to a larger and diverse target domain with more classes. For addressing this specific open set domain adaptation setting, prior research introduces a domain adversarial model that uses a fixed threshold for distinguishing known from unknown target samples and lacks at handling negative transfers. We extend their adversarial model and propose a novel adversarial domain adaptation model with multiple auxiliary classifiers. The proposed multi-classifier structure introduces a weighting module that evaluates distinctive domain characteristics for assigning the target samples with weights which are more representative to whether they are likely to belong to the known and unknown classes to encourage positive transfers during adversarial training and simultaneously reduces the domain gap between the shared classes of the source and target domains. A thorough experimental investigation shows that our proposed method outperforms existing domain adaptation methods on a number of domain adaptation datasets.

Texture is an indispensable property to develop many vision based autonomous applications. Compared to colour, feature dimension in a local texture descriptor is quite large as dense texture features need to represent the distribution of pixel intensities in the neighbourhood of each pixel. Large dimensional features require additional time for further processing that often restrict real-time applications. In this paper, a robust local texture descriptor is enhanced by reducing feature dimension by three folds without compromising the accuracy in region-based image segmentation applications. Reduction in feature dimension is achieved by exploiting the mean of neighbourhood pixel intensities radially along lines across a certain radius, which eliminates the need for sampling intensity distribution at three scales. Both the results of benchmark metrics and computational time are promising when the enhanced texture feature is used in a region-based hierarchical segmentation algorithm, a recent state-of-the-art technique.

Convolutional Neural Networks are highly effective for image classification. However, it is still vulnerable to image distortion. Even a small amount of noise or blur can severely hamper the performance of these CNNs. Most work in the literature strives to mitigate this problem simply by fine-tuning a pre-trained CNN on mutually exclusive or a union set of distorted training data. This iterative fine-tuning process with all known types of distortion is exhaustive and the network struggles to handle unseen distortions. In this work, we propose distortion robust DCT-Net, a Discrete Cosine Transform based module integrated into a deep network which is built on top of VGG16 [1]. Unlike other works in the literature, DCT-Net is "blind" to the distortion type and level in an image both during training and testing. The DCT-Net is trained only once and applied in a more generic situation without further retraining. We also extend the idea of dropout and present a training adaptive version of the same. We evaluate our proposed DCT-Net on a number of benchmark datasets. Our experimental results show that once trained, DCT-Net not only generalizes well to a variety of unseen distortions but also outperforms other comparable networks in the literature.

Multimodal image registration (MMIR) is the alignment of contents in images captured from different sensors or instruments. MMIR is important in medical applications as it enables the visualization of the complementary contents in biomedical microscopic images. The registration for such images can be challenging as the structures of their contents are usually only partially similar. Thus in this paper, we propose a new method to maximize the structural similarity of the contents in such images by utilizing intensity relationships among Red-Green-Blue color channels. Our experimental results will demonstrate that our proposed method substantially improves the accuracy of registering such images as compared to the state-of-the-art methods.

As texture information among pixels can be effectively represented using Local binary patterns (LBPs), image descriptors built using LBPs or its variants have been frequently used for various image analysis applications, e.g. medical image and texture image classification and retrieval. However, neither LBP nor any of its existing variants can be used to build descriptors for classifying multimodal images effectively. This is because the same object when captured in different modalities may result in opposite pixel intensity in some corresponding parts of the images, which in turn will cause their descriptors to be very different. To solve this problem, we propose a novel modality invariant texture descriptor which is built by modifying the standard procedure for building LBP. In this paper, we explain how the proposed descriptor can be built efficiently. We also demonstrate empirically that compared to all the state of the art LBP-based descriptors, the proposed descriptor achieves better accuracy for classifying multimodal images.

This paper introduces a new ensemble approach, Feature-Subspace Aggregating (Feating), which builds local models instead of global models. Feating is a generic ensemble approach that can enhance the predictive performance of both stable and unstable learners. In contrast, most existing ensemble approaches can improve the predictive performance of unstable learners only. Our analysis shows that the new approach reduces the execution time to generate a model in an ensemble through an increased level of localisation in Feating. Our empirical evaluation shows that Feating performs significantly better than Boosting, Random Subspace and Bagging in terms of predictive accuracy, when a stable learner SVM is used as the base learner. The speed up achieved by Feating makes feasible SVM ensembles that would otherwise be infeasible for large data sets. When SVM is the preferred base learner, we show that Feating SVM performs better than Boosting decision trees and Random Forests. We further demonstrate that Feating also substantially reduces the error of another stable learner, k-nearest neighbour, and an unstable learner, decision tree.

One common approach in swarm-based clustering is to use agents to create a set of clusters on a two-dimensional grid, and then use an existing clustering method to retrieve the clusters on the grid. The second step, which we call grid-cluster retrieval, is an essential step to obtain an explicit partitioning of data. In this study, we highlight the issues in grid-cluster retrievals commonly neglected by researchers, and demonstrate the non-trivial difficulties involved. To tackle the issues, we then evaluate three methods: K-means, hierarchical clustering (Weighted Single-link) and density-based clustering (DBScan). Among the three methods, DBScan is the only method which has not been previously used for grid-cluster retrievals, yet it is shown to be the most suitable method in terms of effectiveness and efficiency.

In this paper, we remove the ant-metaphor from ant-based clustering using a randomised partitioning method followed by an agglomerative clustering procedure. While our model only adopts part of the ant-based heuristics, it has produced results that are comparable to the ant-based model. Our approach is based on the fact that one ant can produce the same results as many ants in the models that we have studied, and these models function like stochastic sampling algorithms. In addition, we introduce a schedule to terminate the clustering process before the maximum number of iterations has been reached. We also improve the runtime stability of our model with respect to changes in the structures of the data sets.

Vector quantization histogram (VQH) is a content-based image retrieval technique that has high retrieval effectiveness. However, as VQH's image indexing process is highly computationally intensive, it is not ideal as an online system. We propose an efficient image encoding technique to address this issue. Our experimental results show the proposed technique greatly improves the retrieval efficiency of VQH and it also has little adverse effects on VQH's retrieval effectiveness.

BACKGROUND: Due to the large number of genes in a typical microarray dataset, feature selection looks set to play an important role in reducing noise and computational cost in gene expression-based tissue classification while improving accuracy at the same time. Surprisingly, this does not appear to be the case for all multiclass microarray datasets. The reason is that many feature selection techniques applied on microarray datasets are either rank-based and hence do not take into account correlations between genes, or are wrapper-based, which require high computational cost, and often yield difficult-to-reproduce results. In studies where correlations between genes are considered, attempts to establish the merit of the proposed techniques are hampered by evaluation procedures which are less than meticulous, resulting in overly optimistic estimates of accuracy. RESULTS: We present two realistically evaluated correlation-based feature selection techniques which incorporate, in addition to the two existing criteria involved in forming a predictor set (relevance and redundancy), a third criterion called the degree of differential prioritization (DDP). DDP functions as a parameter to strike the balance between relevance and redundancy, providing our techniques with the novel ability to differentially prioritize the optimization of relevance against redundancy (and vice versa). This ability proves useful in producing optimal classification accuracy while using reasonably small predictor set sizes for nine well-known multiclass microarray datasets. CONCLUSION: For multiclass microarray datasets, especially the GCM and NCI60 datasets, DDP enables our filter-based techniques to produce accuracies better than those reported in previous studies which employed similarly realistic evaluation procedures.