Pattern Recognition Letters

• Reversible adversarial attack can prevent illegal image access while ensure legal use.
• Embed adversarial perturbations of Y channel into UV channel by reversible data hiding.
• Reduce adversarial perturbations by class activation map ...

Recent research on using adversarial perturbation to prevent malicious models from accessing image data has led to the corruption of image data, making images useless in other fields, especially in digital forensics. To prevent malicious models ...

• Adaptively allocate computing resource to input regions for better network inference.
• Patch sampling algorithm reduces redundant calculation costs in overlapping regions.
• Comparable performance is achieved on two datasets by ...

Display Omitted

In this paper, we propose an entropy-dynamic resolution detection (EDRdet) method for object detection in aerial images. Most conventional object detection methods usually detect each region in aerial images directly with a fixed resolution, so ...

• A novel WDPNet is proposed to effectively solve the smooth problem of SR image details with novel L2HID and DPE mechanisms.
• Low- and high-frequency branches recover low-frequency structure and high-frequency details, respectively.
• ...

Single image super-resolution (SR) is an important topic in computer vision because of its ability to generate high-resolution (HR) images. Traditional SR methods do not pay attention to high-frequency detail perception in the reconstruction ...

• Immersive audio experiences are knowing a growing interest.
• 3D audio is often acquired through first order Ambisonics microphones.
• Quaternion algebra is appropriate to process quaternion Ambisonics signals.
• Dual quaternion ...

Spatial audio methods are gaining a growing interest due to the spread of immersive audio experiences and applications, such as virtual and augmented reality. For these purposes, 3D audio signals are often acquired through arrays of Ambisonics ...

• Deep neural networks always require more labelled data to be trained.
• Annotating is time-consuming, learning on a limited amount of data becomes necessary.
• The data to annotate must be correctly chosen to optimize performance.
• ...

Deep neural networks are becoming increasingly powerful and large and always require more labelled data to be trained. However, since annotating data is time-consuming, it is now necessary to develop systems that show good performance while ...

• Face-specific depth enhancement to refine the low-quality depth.
• Iterative inter-modal feature interaction to fully exploit complementary information.
• Feature re-calibration and weighted complementary feature aggregation to ...

Existing RGB-based 2D face recognition approaches are sensitive to facial variations, posture, occlusions, and illumination. Current depth-based methods have been proved to alleviate the above sensitivity by introducing geometric information but ...

• The feature pyramids obtained by the top-down and bottom-up combinations are integrated into one feature map.
• The integrated feature map is input into a self-attention module to yield the learnt feature map.
• Mutual combinations ...

In this paper, we propose a feature enhancement method based on multi-scale self-attention, mainly including a multi-scale feature combination module and a self-attention module. The multi-scale feature combination module integrates the multi-...

• Introduce jigsaw puzzle solving auxiliary loss into vision transformer-based models.
• Removing positional embeddings, randomly masking patches as techniques.
• Improve vision transformers’ generalization on large-scale image ...

The success of Vision Transformer (ViT) in various computer vision tasks has promoted the ever-increasing prevalence of this convolution-free network. The fact that ViT works on image patches makes it potentially relevant to the problem of jigsaw ...

• The FACS-GCN is built to model the inherent AU relations referring to FACS.
• The DLR-GCN is built to complement individual differences ignored by FACS-GCN.
• The DGCN is proposed to integrate two types factors impacting AU ...

Facial Action Unit (AU) recognition is a challenging problem, where the subtle muscle movement brings diverse AU representations. Recently, AU relations are utilized to assist AU recognition and improve the understanding of AUs. Nevertheless, ...

• A “one-shot” pruning method based on the redundancy of filters for CNNs is proposed.
• Graphs are established to represent the similarity relations between feature maps.
• The proposed method is tested on VGGNet, ResNet, and DenseNet ...

Display Omitted

The vast majority of repetitive pruning and retraining techniques used on CNNs require multi-stage optimization, which undermines the potential computing savings from pruning. The similarity relationship between the output feature maps of the ...

The bulk of the documents that affect our lives are digital or born digital. Our laborious investigations of layout, script, font and graphics, are turning into mere exercises with little influence on pursuits outside the Document Analysis and ...

Display Omitted

• Propose a dual-branch pyramid network (DBPNet) based on differences in feature distributions of document images.
• Design a text edge restoration module based on laplacian pyramid structure to improve the edge details.
• ...

Convolutional neural networks (CNN), aiming to preserve the structural and texture in- formation lost in the initial low-resolution (LR) images, has been widely used to improve the quality of LR images. Most of the existing super-resolution ...

• We fill the gap in the literature by investigating image classification using multiscale superpixels.
• WaveMesh, a novel wavelet-based superpixeling algorithm is proposed.
• In WaveMesh, the number and size of superpixels are computed ...

Prior studies using graph neural networks (GNNs) for image classification have focused on graphs generated from a regular grid of pixels or similar-sized superpixels. In the latter, a single target number of superpixels is defined for an entire ...

We present SIT-SR 3D, a novel self-supervised method for 3D single image super-resolution (SISR). Scaling 2D SISR networks to 3D SISR requires code redesign, high computing resources, and 3D ground-truth. However, we circumvent this by (1) using ...

• Cairo pattern is a tiling by pentagons and is a dual of a semi-regular tiling.
• Four orientations of pentagons resulted as the intersection of two hexagonal grids.
• There are 4 types of neighbors, thus there are 4 weights depending ...

The Cairo Pattern is an interesting tiling of the plane. It consists of pentagons which completely cover the plane without overlapping. The Cairo Pattern is the dual of a semi-regular grid. It is assumed that one can walk on the pentagons such ...

• The proposed model is a modernized version of Edit-Unsup-TS that uses masked language models.
• The idea of fine-tuning BERT on simpler language is explored.
• It improves performance with a lower amount of training data.

Word suggestion in unsupervised sentence simplification aims to replace complex words of a given sentence with their simpler alternatives. This is mostly done without considering their context within the input sentence. In this paper, we propose ...

• The proposed method adopts a concept motivated by human cognition, looking through objects at different intervals.
• The proposed method captures local fluctuations and global trends in data using different intervals.
• The proposed ...

This paper presents a new multiscale transformation for statistical analysis of one-dimensional data such as time series under the concept of the scale-space approach. The proposed method uses regular observations (eye scanning) with a range of ...

• A generator is learned to augment semantic attributes to solve the projection problem of “many to one”.
• The reconstructed features and the original features are concatenated to train the classifier.
• The proposed method can achieve ...

Generalized Zero-Shot Learning (GZSL) has become an important research due to its powerful ability of recognizing unseen objects. Generative methods, converting conventional GZSL to fully supervised learning, can achieve competing performance, ...

• Data Integration of two or more heterogeneous databases.
• Syntactic and semantic analysis of textual data.
• Semi-automatic process.

Display Omitted

One of the most difficult issues today, is the integration of data from various sources. Thus, it arises the need of automatic Data Integration (DI) methods. However, in the literature there are fully automatic or semi-automatic DI techniques, ...

• A zero-shot cross-dataset transfer protocol.
• Competitive cross-dataset results.
• The leverage of a pipeline built on top of a pre-trained backbone.

Display Omitted

Smartphones contain personal and private data to be protected, such as everyday communications or bank accounts. Several biometric techniques have been developed to unlock smartphones, among which ear biometrics represents a natural and promising ...

• A novel lossis proposed for few-shot RE to enlarge the inter-class margin.
• Two-step training improves the model performance in few-shot RE with NOTA detection.
• A SOTA sentence-pair model for few-shot RE with NOTA detection.

Few-shot relation extraction aims to solve the problem of insufficient annotated data in relation extraction tasks. Through the comparison between samples, few-shot relation extraction achieves lower-cost relation classification. However, most ...

• We proposed a deep-learning-based model for perceptual speech quality assessment.
• We presented an application of this model as perceptual regularization for speech enhancement.
• Experimental results show significant improvement in ...

Many speech enhancement methods require perceptual quality metrics for evaluation. The “holy grail” of perceptual speech quality assessment is human subjective ratings, known as the mean opinion score. However, acquiring human ratings is time-...

• Challenging machine learning problems are unlikely to have trivial solutions.
• Solutions from low-capacity models are likely shortcuts that won’t generalize.
• One inductive bias for robust generalization is to avoid overly simple ...

Despite their impressive performance in object recognition and other tasks under standard testing conditions, deep networks often fail to generalize to out-of-distribution (o.o.d.) samples. One cause for this shortcoming is that modern ...