Search

Regarding the interior non-local correlations of image content, the intra predictions in image/video coding standards have still not exploited them efficiently. In this article, we propose an intra coding method that leverages the geometric information extracted from the image content to strengthen the coding performance. During the process of image capturing, if the imaging plane of the camera is not fronto-parallel to the surface of an object in reality, the repetitive patterns on the object’s surface in real world will appear with a scale shift in the image domain. Thus, the corresponding potential non-local redundancy cannot be removed through the direct utilization of intra block copy and its variants. To address this problem, we begin with a theoretic analysis into the perspective transformation matrix and derive the underlying geometric information that causes the scale shift in the image domain; afterwards, we propose an intra coding framework based on the geometric information to alleviate the scale shift. It is essential for the coding framework to realize block matching in the rectified domain. Our framework mainly consists of two components, i.e., the intra block copy in the rectified domain through planar perspective transformation to obtain a more accurate prediction for the current coding block and the non-local post-processing filtering in the rectified domain after decompression to achieve a better quality of the reconstructed image. The experimental results show that our proposed method can achieve as high as 18% and an average of 5.0% bit-rate saving on the common dataset Urban100 compared to the HEVC reference software HM-SCC-extension with intra block copy and non-local post-processing filtering.

The ability to hide plays a vital role in effective secret communication. This is achieved by hiding information (Steganography). The science of concealing information is the science of concealing information in other information so that it appears that hidden information is not something to the human eye. There are many ways to hide information within an image, audio / video, document, etc. But hiding information in pictures has its own characteristics and is the most popular among others. This paper provides a review of several methods, such as image field and conversion field algorithms available to implement image information hiding (Steganography). In this paper, high-capacity information hidings schemes are analyzed for different file formats. Secret communication is done before Password encryption to protect information. The intended recipient will decrypt the information using this password.

This chapter presents methods that allow us to analyse or modify the geometric properties of an image. Our attention, first, is on techniques for filtering images to remove noise or to enhance geometric detail. We will then look at means by which we can characterise geometric properties like texture, and processes that allow us to analyse objects and shapes that appear in imagery.

Human Activity Recognition (HAR) is the most popular research area in the pervasive computing field in recent years. Sensor data plays a vital role in identifying several human actions. Convolutional Neural Networks (CNNs) have now become the most recent technique in the computer vision phenomenon, but still, it is premature to use CNN for sensor data, particularly in ubiquitous and wearable computing.Deep CNN requires huge dataset and models which increases the computational complexity. Transfer learning that uses the pre trained CNNwith fine tuning is the better alternative to reduce the training cost.In this paper, we have proposed the idea of transforming the raw accelerometer and gyroscope sensor data to the visual domain by using our novel activity image creation method (NAICM). Pre-trained CNN (AlexNet) has been used on the converted image domain information. The proposed method is evaluated on several online available human activity recognition dataset. The results show that the proposed novel activity image creation method (NAICM) has successfully created the activity images with a classification accuracy of 98.36% using pre trained CNN.

This study deals with a tracking problem for hidden human targets using time-division multiple-input-multiple-output through-wall imaging radar (TWIR). An efficient image-domain tracking algorithm is proposed. Specifically, the authors first utilise back-projection algorithm and the phase coherence factor (PCF) to obtain multi-frame high quality images. Then a tracking algorithm via amplitude-distribution-based particle filter is proposed. Experimental data validates that this algorithm has a commendable effectiveness for hidden target tracking.

A lot of research attentions have been paid to image emotion analysis in recent years. Meanwhile, as convolutional neural networks (CNNs) have made great successful in computer vision, many researchers start to employ CNN to discriminate image emotions. However, the training procedure of CNNs depends on sufficient labeled data. Therefore, a CNN is hard to perform well in an image domain with scant labeled information. In this paper, we propose a deep transfer learning method for image emotion analysis. The method can leverage rich emotion knowledge from a source domain to the target domain. Our method reduces both marginal and joint domain distribution discrepancies at fully-connected layers. Through this way, we can effectively extract more transferable features and advance the performance of CNNs on poor-label emotion-image domains.

We present Robbie : a general work-flow for the detection and characterization of radio variability and transient events in the image domain. Robbie is designed to work in a batch processing paradigm with a modular design so that components can be swapped out or upgraded to adapt to different input data, whilst retaining a consistent and coherent methodological approach. Robbie is based on commonly used and open software, and is encapsulated in a Makefile to aid portability and reproducibility. In this paper we describe the methodology behind Robbie , and demonstrate its use on real and simulated data. Robbie is available on GitHub .

Coronavirus pandemic is burdening healthcare systems around the world to the full capacity they can accommodate. There is an overwhelming need to find a treatment for this virus as early as possible. Computer algorithms and deep learning can participate positively by finding a potential treatment for SARS-CoV-2. In this paper, a deep learning model and machine learning methods for the classification of potential coronavirus treatments on a single human cell will be presented. The dataset selected in this work is a subset of the publicly online datasets available on RxRx.ai. The objective of this research is to automatically classify a single human cell according to the treatment type and the treatment concentration level. A DCNN model and a methodology are proposed throughout this work. The methodical idea is to convert the numerical features from the original dataset to the image domain and then fed them up into a DCNN model. The proposed DCNN model consists of three convolutional layers, three ReLU layers, three pooling layers, and two fully connected layers. The experimental results show that the proposed DCNN model for treatment classification (32 classes) achieved 98.05% in testing accuracy if it is compared with classical machine learning such as support vector machine, decision tree, and ensemble. In treatment concentration level prediction, the classical machine learning (ensemble) algorithm achieved 98.5% in testing accuracy while the proposed DCNN model achieved 98.2%. The performance metrics strengthen the obtained results from the conducted experiments for the accuracy of treatment classification and treatment concentration level prediction.

Magnetic resonance imaging (MRI) is a dynamic and safe imaging technique in medical imaging. Recently, parallel MRI (pMRI) is widely used for accelerating conventional MRI. Both frequency and image domain-based reconstructions are the most attractive methods for generating the image from multi-channel k-space data. Compressed sensing (CS) is a recently used procedure to reduce the acquisition time of conventional MRI. This reduction is achieved by taking fewer measurements from the fully sampled k-space data. Therefore, applying the CS technique in pMRI is the most emerging way for further improving the acquisition time that is a tremendous research interest. However, as the phase encoding plane may be perpendicular or parallel to the coil elements plane, finding the exact domain for CS in pMRI reconstruction is a major challenging issue. In this work, the application of the CS technique in pMRI in both domains is investigated. Later some widely used methodologies are presented as the nonlinear reconstruction algorithm of CS in pMRI. Finally, a discussion is performed based on CS in pMRI to perceive the reality of different reconstruction algorithms at a glance for finding preferred methodologies.

Dual-energy computer tomography (DECT) has high application prospects in distinguishing and quantifying materials. However, DECT requires higher hardware cost and higher radiation dose than single-energy CT imaging. In this paper, we have developed a cascaded network method to realize DECT imaging through SECT images, then get the basis material decomposition images. Specifically, we design the mapping convolutional neural network and the material decomposition U-Net to realize the mapping of low- to high-energy images and material decomposition, respectively. To verify the feasibility of the proposed method, we extracted 1442 cranial cavity slice images of 5 patients for experiments. The qualitative and quantitative results show that the proposed method can achieve high-quality DECT imaging of single energy data and high-precision basis material decomposition.

PurposeDeep learning (DL) models have rapidly become a popular and cost-effective tool for image classification within oncology. A major limitation of DL models is output instability, as small perturbations in input data can dramatically alter model output. The purpose of the study is to investigate the robustness of DL models in the oncologic image domain through the application of adversarial images: manipulated images with small pixel-level perturbations designed to assess the stability of DL models.Experimental DesignWe examined the impact of adversarial images on the classification accuracies of DL models trained to classify cancerous lesions across three common oncologic imaging modalities (CT, mammogram, and MRI). The CT model was trained to classify malignant lung nodules using the LIDC dataset. The mammogram model was trained to classify malignant breast lesions using the DDSM dataset. The MRI model was trained to classify brain metastases using an institutional dataset. We also explored the utility of an iterative adversarial training approach to improve the stability of DL models to small pixel-level changes.ResultsOncologic images showed instability with small pixel-level changes. A pixel-level of perturbation of .004 resulted in a majority of oncologic images to be misclassified by their respective DL models (CT 25.64%, mammogram 23.93%, MRI 6.36%). Adversarial training mitigated improved the stability and robustness of DL models trained on oncologic images compared to naive models [(CT 67.72% vs 26.92%), mammogram (63.39% vs 27.68%), MRI (87.20% vs 24.32%)].ConclusionsDL models naively trained on oncologic images exhibited dramatic instability to small pixel-level changes resulting in substantial decreases in accuracy. Adversarial training techniques improved the stability and robustness of DL models to such pixel-level changes. Prior to clinical implementation, adversarial training should be considered to proposed DL models to improve overall performance and safety.

Language-driven image editing can significantly save the laborious image editing work and be friendly to the photography novice. However, most similar work can only deal with a specific image domain or can only do global retouching. To solve this new task, we first present a new language-driven image editing dataset that supports both local and global editing with editing operation and mask annotations. Besides, we also propose a baseline method that fully utilizes the annotation to solve this problem. Our new method treats each editing operation as a submodule and can automatically predict operation parameters. Not only performing well on challenging user data, but such an approach is also highly interpretable. We believe our work, including both the benchmark and the baseline, will advance the image editing area towards a more general and free-form level.

To highlight the challenges of achieving representation disentanglement for text domain in an unsupervised setting, in this paper we select a representative set of successfully applied models from the image domain. We evaluate these models on 6 disentanglement metrics, as well as on downstream classification tasks and homotopy. To facilitate the evaluation, we propose two synthetic datasets with known generative factors. Our experiments highlight the existing gap in the text domain and illustrate that certain elements such as representation sparsity (as an inductive bias), or representation coupling with the decoder could impact disentanglement. To the best of our knowledge, our work is the first attempt on the intersection of unsupervised representation disentanglement and text, and provides the experimental framework and datasets for examining future developments in this direction., Comment: Accepted to RepL4NLP 2021

Transfer learning enables to re-use knowledge learned on a source task to help learning a target task. A simple form of transfer learning is common in current state-of-the-art computer vision models, i.e. pre-training a model for image classification on the ILSVRC dataset, and then fine-tune on any target task. However, previous systematic studies of transfer learning have been limited and the circumstances in which it is expected to work are not fully understood. In this paper we carry out an extensive experimental exploration of transfer learning across vastly different image domains (consumer photos, autonomous driving, aerial imagery, underwater, indoor scenes, synthetic, close-ups) and task types (semantic segmentation, object detection, depth estimation, keypoint detection). Importantly, these are all complex, structured output tasks types relevant to modern computer vision applications. In total we carry out over 2000 transfer learning experiments, including many where the source and target come from different image domains, task types, or both. We systematically analyze these experiments to understand the impact of image domain, task type, and dataset size on transfer learning performance. Our study leads to several insights and concrete recommendations: (1) for most tasks there exists a source which significantly outperforms ILSVRC'12 pre-training; (2) the image domain is the most important factor for achieving positive transfer; (3) the source dataset should \emph{include} the image domain of the target dataset to achieve best results; (4) at the same time, we observe only small negative effects when the image domain of the source task is much broader than that of the target; (5) transfer across task types can be beneficial, but its success is heavily dependent on both the source and target task types., Comment: Accepted for future publication in TPAMI

International audience; In this paper, we demonstrate that algebraic topology can be used to perform 2D+t object detection. After the construction of a topological complex for a 2D+t image sequence, we build a nested sequence of cell complexes on which relative persistent homology is computed. The relative homology adds to “absolute” homology the computation of classes related to the first and last frames of the sequence. By identifying 2D chains with large life spans, the most persistent classes are extracted. This allows for the identification of the interesting parts in a sequence and for the detection of the movement of objects despite continuous deformations in the image domain. The results obtained on a synthetic image and on two real biomedical images with moving vesicles recorded by a quantitative phase time‐lapse technique show the potential of this method. Comparing the method with a newly developed tracking tool proves that the strength of this method is its independence from prior parameters.

Inter-subject registration of cardiac images is a vital yet challenging task due to the large deformations influenced by the cardiac cycle and respiration. Various intensity-based cardiac registration methods have already been proposed, but such methods utilize intensity information over the entire image domain and are thus computationally expensive. In this work, we propose a novel pipeline for fast registration of cardiac MR images that relies on shape priors and the strategic location of surface-approximating landmarks. Our holistic approach to cardiac registration requires minimal user input. It also reduces the computational runtime by \(60\%\) on average, which amounts to an 11-min speedup in runtime. Most importantly, the resulting Dice similarity coefficients are comparable to those from a widely used elastic registration method.

Coronavirus pandemic is burdening healthcare systems around the world to the full capacity they can accommodate. There is an overwhelming need to find a treatment for this virus as early as possible. Computer algorithms and deep learning can participate positively by finding a potential treatment for SARS-CoV-2. In this paper, a deep learning model and machine learning methods for the classification of potential coronavirus treatments on a single human cell will be presented. The dataset selected in this work is a subset of the publicly online datasets available on RxRx.ai. The objective of this research is to automatically classify a single human cell according to the treatment type and the treatment concentration level. A DCNN model and a methodology are proposed throughout this work. The methodical idea is to convert the numerical features from the original dataset to the image domain and then fed them up into a DCNN model. The proposed DCNN model consists of three convolutional layers, three ReLU layers, three pooling layers, and two fully connected layers. The experimental results show that the proposed DCNN model for treatment classification (32 classes) achieved 98.05% in testing accuracy if it is compared with classical machine learning such as support vector machine, decision tree, and ensemble. In treatment concentration level prediction, the classical machine learning (ensemble) algorithm achieved 98.5% in testing accuracy while the proposed DCNN model achieved 98.2%. The performance metrics strengthen the obtained results from the conducted experiments for the accuracy of treatment classification and treatment concentration level prediction.

This paper mainly focuses on synthetic aperture radar (SAR) ground moving target imaging. Although there exists many excellent SAR moving target imaging algorithms, two issues, the maneuverability of the SAR platform and the type of data used for moving target imaging, are not discussed by most of them. Thus, a novel SAR image domain-ground moving target imaging method is proposed to preliminarily handle the aforementioned two issues. The method proposed contains two main steps. The first one is the pre-imaging of the raw data, and the second one is focusing the ground moving target's image data by a proposed one-dimensional parameter traversal approach. Numerical experiments are finally presented to verify the effectiveness of the proposed ground moving target imaging method.

Leveraging the power of deep neural networks, single-person pose estimation has made substantial progress throughout the last years. More recently, multi-person pose estimation has also become of growing importance, mainly driven by the high demand for reliable video surveillance systems in public security. To keep up with these demands, certain efforts have been made to improve the performance of such systems, which is yet limited by the insufficient amount of available training data. This work addresses this lack of labeled data: by diminishing the often faced problem of domain shift between synthetic images from computer game graphics engines and real world data, annotated training data shall be provided at zero labeling-cost. To this end, generative adversarial networks are applied as domain adaption framework, adapting the data of a novel synthetic pose estimation dataset to several real world target domains. State-of-the-art domain adaption methods are extended to meet the important requirement of exact content preservation between synthetic and adapted images. Experiments, that are subsequently conducted, indicate the improved suitability of the adapted data as human pose estimators trained on this data outperform those which are trained on purely synthetic images.

The linear structures in synthetic aperture radar (SAR) images can provide important geometric information regarding the illuminated objects. However, the linear structures of various objects sometimes disappear in traditional SAR images, which severely affects the application in automated scene analysis and information extraction techniques. This paper proposes a parametric SAR image recovery method for linear structures of extended targets. By extracting the spatial phase distribution feature in image domain, the proposed method is used to identify the endpoints among all of the scattered points in SAR images and reconstruct the disappeared linear structures based on the scattering model. The method can be generally divided into three procedures: endpoints and point targets classification, linear structures recognition, and linear structures reconstruction. In the first step, the endpoints of linear structures and point targets are classified by setting a threshold related to the spatial phase distribution feature. Afterwards, the linear structures recognition method is used to determine which two endpoints can be formed into a linear structure. Finally, the parametric scattering models are used to reconstruct the disappeared linear structures. Experiments are conducted on both computer simulations and the data that were acquired by microwave anechoic chamber experiment, tower crane radar experiment, and unmanned vehicle radar experiment in order to validate the effectiveness and robustness of the proposed method.

The aim of the study was to evaluate the postoperative quality of life (QoL) of patients who underwent minimally invasive repair of pectus excavatum (MIRPE) with a newly designed bar and bar stabilizers.We conducted a prospective randomized study in which patients were operated either with standard perpendicular stabilizers (control group) or with the newly designed oblique stabilizers (intervention group). All patients were evaluated 6 months after the operation with the Pectus Excavatum Evaluation Questionnaire (PEEQ).There were 16 patients in the control group and 14 in the intervention group. Mean age was 17 (SD: 3.3, range 14-27) years. There were no demographic differences between groups. Two patients in the control group and one in the intervention group were repaired with two bars instead of one. There was one reoperation in each group. There was a significant difference between the pre- and postoperative scores, in both groups, in the patient body image domain (control group: 9.5 to 3; p 0.01; intervention group 10 to 3; p 0.01), as well as in the psychosocial domain (control group: 13.5 to 24, p 0.01; intervention group: 15 to 24, p 0.01). With regards to the patients' perception of physical difficulties before and after MIRPE, the difference between pre- and postoperative scores was greater in the intervention group (8 to 12, p 0.01) than in the control group (10 to 11, p = 0.04). The mean length of stay was 4.5 and 5 days in the intervention group and the control group, respectively.Our study showed that patients who underwent MIRPE with the newly designed bars and stabilizers had non-inferior outcomes than patients reported in the literature who underwent MIRPE with standard bars and stabilizers. We found slightly better outcomes in patients in the intervention group compared to the control group, but larger studies will be needed to confirm if those differences are statistically significant.II.

This research focuses on the problem of shape analysis on the 2D image domain. State of the art methods like convolution neural networks typically rely on the richness of texture information in images for inference. However, when the underlying object is better described as a shape, these methods tend to suffer. Here, we aim to bridge this gap by proposing a method to analyze shapes on images. The driving idea is to learn features defined on just a few point samples extracted from image super-pixels. A dynamic graph CNN (i.e., a neural net producing a different graph at each layer) is trained and used as the learning engine in a classification task. Our first set of experiments is tested on the 10-digit class MNIST dataset benchmark where we find that the proposed method performs better than others on small datasets. This altogether shows a promising direction for the analysis of more complex shapes fundamental in classification and retrieval of documents of scientific nature.

This paper presents a memorability based image-to-image translation technique to make an image more memorable while retaining its high-level contents. Conventionally, the image-to-image translation task aims to learn the mapping between images of two different domains using a set of aligned image pairs. However, dataset having such one-to-one mapping is not available for memorability based image-to-image translation. Therefore, the aim of the proposed task is defined to learn the mapping F: I → I' between two image domains I and I'. Here, I corresponds to input image domain and I' is the unknown image domain containing the modified version of the input images. Also, every image in I' is more memorable than its corresponding image in I. Therefore, the proposed task is achieved by developing a deep learning based method to learn the mapping F: I→ I' using mean-squared error and memorability loss between I and F(I). The experimental results showed that the proposed approach increases the memorability of the given image better than the state-of-the-art image-to-image translation techniques.

A new interferometric inverse synthetic aperture radar (InISAR) imaging method based on a L shape three-antenna interferometric radar geometry and dominant-scatterer phase focusing is presented in this paper. Under this InISAR model, the distance of the target in the baseline direction introduces additional interferometric phase offset, which can cause phase ambiguity, so the interferometric phase must be corrected before interferometric processing. Firstly, a strong scattering point is searched for as a reference point in the image domain, and then the phase of the scattering point is used as the reference phase for phase focusing on each ISAR image respectively to unwrap the interferometric phase, so that the 3D coordinates of the target can be correctly retrieved. Simulation results confirm the effectiveness of the method.

Convolutional neural networks have shown impressive performance in the medical image domain, but medical experts are somewhat skeptical in their predictions since the features are not directly graspable. We are looking into one of the technical challenges, namely the explainability of the results, to try to find out some demonstration of the regions deemed abnormal by deep learning. Following the trend of heatmaps indicating which local morphology changes would modify the predictions, we are trying to verify the facilitation of the clinical understanding in the eyes of the ophthalmologist.

For multiple object path tracking we integrate entropy background subtraction method and CAMShift algorithm. Initially claussius entropy theory is used to transform each pixel in the image domain into entropy domain and obtain its energy level. Later we use entropy background subtraction algorithm to detect moving object region in each frame. To improve robustness in the condition where objects are of different color, object colors are same as to background’s colors. localization of object is obtained by choosing each objective region. This approach can automatically calibrate moving vehicles in traffic video and achieve multi-target tracking by using a multi-tracker of CAMShift algorithm. The proposed system is also capable of tracing paths of moving vehicles. The results and analysis demonstrates that the methods used in the paper finds solution for automatic multi-object tracking problems in video sequence efficiently.

Adversarial examples (AEs) are crafted by adding human-imperceptible perturbations to inputs such that a machine-learning based classifier incorrectly labels them. They have become a severe threat to the trustworthiness of machine learning. While AEs in the image domain have been well studied, audio AEs are less investigated. Recently, multiple techniques are proposed to generate audio AEs, which makes countermeasures against them urgent. Our experiments show that, given an audio AE, the transcription results by Automatic Speech Recognition (ASR) systems differ significantly (that is, poor transferability), as different ASR systems use different architectures, parameters, and training datasets. Based on this fact and inspired by Multiversion Programming, we propose a novel audio AE detection approach MVP-Ears, which utilizes the diverse off-the-shelf ASRs to determine whether an audio is an AE. We build the largest audio AE dataset to our knowledge, and the evaluation shows that the detection accuracy reaches 99.88%. While transferable audio AEs are difficult to generate at this moment, they may become a reality in future. We further adapt the idea above to proactively train the detection system for coping with transferable audio AEs. Thus, the proactive detection system is one giant step ahead of attackers working on transferable AEs.

Artefacts caused by the presence of metallic implants and prosthesis appear as dark and bright streaks in computed tomography (CT) images, that obscure the information about underlying anatomical structures. These phenomena can severely degrade the image quality and hinder the correct diagnostic interpretation. Although many techniques for the reduction of metal artefacts have been proposed in literature, their effectiveness is still limited. In this paper, an application of a convolutional neural networks (CNN) to the problem of metal artefact reduction (MAR) in the image domain is investigated. Experimental results show that image-domain CNN can substantially suppresses streaking artefacts in the reconstructed images.

In image-to-image translation the goal is to learn a mapping from one image domain to another. In the case of supervised approaches the mapping is learned from paired samples. However, collecting large sets of image pairs is often either prohibitively expensive or not possible. As a result, in recent years more attention has been given to techniques that learn the mapping from unpaired sets. In our work, we show that injecting implicit pairs into unpaired sets strengthens the mapping between the two domains, improves the compatibility of their distributions, and leads to performance boosting of unsupervised techniques by up to 12% across several measurements. The competence of the implicit pairs is further displayed with the use of pseudo-pairs, i.e., paired samples which only approximate a real pair. We demonstrate the effect of the approximated implicit samples on image-to-image translation problems, where such pseudo-pairs may be synthesized in one direction, but not in the other. We further show that pseudo-pairs are significantly more effective as implicit pairs in an unpaired setting, than directly using them explicitly in a paired setting.

Vessel segmentation has always been a considerable challenge task due to the presence of varying thickness of the vessels and weak contrasts of medical image intensities. In this paper, an effective method is proposed, which consists of four steps. Firstly, the input images are converted into gray ones with predetermined weightings aiming at increasing image contrast if they are colorful. Secondly, the image intensities are expanded from regions of interest to the whole image domain with a mirroring operation to avoid introducing undesired boundaries by image filtering operations existing in the next step. Thirdly, an improved multi-scale enhancement method inspired by the Frangi filtering is proposed to enhance image contrast between blood vessels and other objects in the image. Finally, an improved level set model is proposed to segment blood vessels from the enhance images and the original gray images. The proposed method has been evaluated on two retinal vessel image repositories, namely, DRIVE and STARE. Experimental results and comparison with 13 existing methods show that the proposed method produces similar segmentation accuracy at the same level with representative methods in the literature. Its effectiveness on segmentation of other type vessels is also discussed at the end of this paper.

Traditional strategies for reconstructing Compressed Sensing Magnetic Resonance Imaging (CS-MRI) may introduce computational redundancy, and deep learning-based methods can significantly reduce reconstruction time and improve restoration quality. However, many recent deep learning-based algorithms lay insufficient attention to spatial frequency information. In this paper, a Structural Oriented Generative Adversarial Network (SOGAN) is proposed aiming at restoring image domain information as well as refining frequency domain during the reconstruction of CS-MRI. Numerical Experiments showed our model’s efficiency and capability for diagnostic purpose.

Fast reconstruction of under-sampled acquisitions has always been a central issue in MRI reconstruction. Recently years has seen multiple studies using deep learning as a de-aliasing framework to restore the aliased image. However, restoration of fine details is still problematic, especially when dealing with noisy image datasets. Sparked by the Fourier transform relationship, this work proposed and tested a new hypothesis: can regularization be directly added in the frequency domain to correct the high-frequency imperfection? To achieve this, discriminative networks are applied in both the image domain and the frequency domain (so-called dual-domain GAN). Evaluation on multiple datasets proved that the dual-domain GAN approach is an effective way to improve the quality of accelerated MR reconstruction.

Multiple instance classification (MIC) is a kind of supervised learning, where data are represented as bags and each bag contains many instances. Training bags are given a label and the system tries to learn how to label unknown bags, without necessarily learning how to label individually each of their instances. In particular, we apply concepts drawn from MIC to the realm of content-based image retrieval, where images are described as bags of visual local descriptors. We introduce several classifiers, according to the different MIC paradigms, and evaluate them experimentally on a real-world dataset, comparing their accuracy and efficiency. © 2019, Springer Nature Switzerland AG.

Magnetic particle imaging is a tracer-based medical imaging technology that is quite promising for the task of imaging vessel structures or blood flows. From this possible application it can be deduced that significant areas of the image domain are related to background, because the tracer material is only inside the vessels and not in the surrounding tissue. From this fact alone it seems promising to detect the background of the image in early stages of the reconstruction process. This paper proposes a multiresolution and segmentation based reconstruction, where the background is detected on a coarse level of the reconstruction with only few degrees of freedom by a Gaussian-mixture model and transferred to finer reconstruction levels.

X-ray images can show great variation in contrast and noise levels. In addition, important subject structures might be superimposed with surgical tools and implants. As medical image datasets tend to be of small size, these image characteristics are often under-represented. For the task of automated, learning-based segmentation of bone structures, this may lead to poor generalization towards unseen images and consequently limits practical application. In this work, we employ various data augmentation techniques that address X-ray-specific image characteristics and evaluate them on lateral projections of the femur bone. We combine those with data and feature normalization strategies that could prove beneficial to this domain. We show that instance normalization is a viable alternative to batch normalization and demonstrate that contrast scaling and the overlay of surgical tools and implants in the image domain can boost the representational capacity of available image data. By employing our best strategy, we can improve the average symmetric surface distance measure by 36:22 %.

Artificial neural networks have been successfully used for many different classification tasks including malware detection and distinguishing between malicious and non-malicious programs. Although artificial neural networks perform very well on these tasks, they are also vulnerable to adversarial examples. An adversarial example is a sample that has minor modifications made to it so that the neural network misclassifies it. Many techniques have been proposed, both for crafting adversarial examples and for hardening neural networks against them. Most previous work was done in the image domain. Some of the attacks have been adopted to work in the malware domain which typically deals with binary feature vectors. In order to better understand the space of adversarial examples in malware classification, we study different approaches of crafting adversarial examples and defense techniques in the malware domain and compare their effectiveness on multiple data sets.

The finger knuckle print (FKP) based personal identification has achieved much interest. One of the disadvantages of the generally used Gabor filter based Competitive Coding method is that it is sensitive to variations caused in image acquisition stage. To handle this problem, researchers have proposed a reconstruction based scheme in original image pixel domain. However, the reconstruction in original image domain may lose much valuable directional information extracted by Gabor filters with different orientations. Compared with the previous methods, in this paper, we propose to perform reconstruction in Gabor filter response domain. For each orientation, we reconstruct the Gabor response of a query image using the responses in the template database. Thus, the directional information contained in the image is fully utilized. Then, we binarize each reconstructed responses for feature coding and matching with the constructed "mask", which considers not only the reconstructive error but also the Gabor response in a specific location. The mask can characterize the significance of a specific location along a specific orientation. Finally, by using a score level adaptive binary fusion rule, the matching distances before and after reconstruction are adaptively fused. The experimental results conducted on the benchmark PolyU FKP database show the performance advantage of the presented method.

Energy spectrum CT can perform spectral imaging of objects. In the clinical and industrial fields, spectral images are often converted into a distribution image of the base material, ie, material decomposition. Traditional image domain material decomposition algorithms are often parsed point by point. This time we try to use the end-to-end AUTOMAP for fitting solution. AUTOMAP can perform image domain mapping and achieve certain effects on MRI image reconstruction, but our attempts at spectral CT material decomposition are not satisfactory.

Person re-identification (re-ID) models trained on one domain often fail to generalize well to another. In our attempt, we present a "learning via translation" framework. In the baseline, we translate the labeled images from source to target domain in an unsupervised manner. We then train re-ID models with the translated images by supervised methods. Yet, being an essential part of this framework, unsupervised image-image translation suffers from the information loss of source-domain labels during translation. Our motivation is two-fold. First, for each image, the discriminative cues contained in its ID label should be maintained after translation. Second, given the fact that two domains have entirely different persons, a translated image should be dissimilar to any of the target IDs. To this end, we propose to preserve two types of unsupervised similarities, 1) self-similarity of an image before and after translation, and 2) domain-dissimilarity of a translated source image and a target image. Both constraints are implemented in the similarity preserving generative adversarial network (SPGAN) which consists of an Siamese network and a CycleGAN. Through domain adaptation experiment, we show that images generated by SPGAN are more suitable for domain adaptation and yield consistent and competitive re-ID accuracy on two large-scale datasets., Comment: Accepted in CVPR 2018

Terror attacks are often targeted towards the civilians gathered in one location (e.g., Boston Marathon bombing). Distinguishing such ’malicious’ scenes from the ’normal’ ones, which are semantically different, is a difficult task as both scenes contain large groups of people with high visual similarity. To overcome the difficulty, previous methods exploited various contextual information, such as language-driven keywords or relevant objects. Although useful, they require additional human effort or dataset. In this paper, we show that using more sophisticated and deeper Convolutional Neural Networks (CNNs) can achieve better classification accuracy even without using any additional information outside the image domain. We have conducted a comparative study where we train and compare seven different CNN architectures (AlexNet, VGG-M, VGG16, GoogLeNet, ResNet- 50, ResNet-101, and ResNet-152). Based on the experimental analyses, we found out that deeper networks typically show better accuracy, and that GoogLeNet is the most favorable among the seven architectures for the task of malicious event classification.

The outdoor RCS measurement technique is crucial for the radar and aeronautical design applications, especially in the face of large objects. Because of the background interference, the multi-bounce effect is unavoidable and will contaminate the measuring result greatly. In this paper, a novel method to eliminate the multi-bounce effect via 3D imaging technique is proposed. Compared with the traditional methods, high-resolution 3D microwave image can separate the desired image from the multi-bounce false images in the image domain by using the difference of the transmission paths. Simulation results show that this method can clearly eliminate the multi-bounce effect of the outdoor RCS measurement.

Conventional amplitude inversion assumes that the migrated image preserves relative-amplitude information. However, illumination effects caused by complex geologic settings, undersampled acquisition geometry, and limited recording aperture pose a challenge to even the most advanced imaging algorithms. In addition, standard depth-migration images can suffer from lack of resolution caused by wavelet stretch, attenuation, and suboptimal deghosting. Least-squares migration (LSM) can mitigate many of these problems and produce better resolved migration images suitable for AVO inversion. However, whether formulated in the data domain or the image domain, LSM is an inversion algorithm and is sensitive to inaccuracies in the source wavelet, velocity model, data preprocessing, and the propagator used. Practical considerations to mitigate these problems under nonideal conditions and cost-reduction strategies differ between the data- and image-domain formulations. The relative merits of each approach are evaluated by using example inversions for complex synthetic models, including free-surface ghost and attenuation effects. When a data-domain implementation of LSM is considered necessary, the image-domain implementation should be considered at the same time, especially when targeting localized reservoir targets under complex overburdens. Application of image-domain least-squares migration on a Gulf of Mexico field data set produces significant improvements in resolution and event continuity in the subsalt target region.

This article presents an introduction to visual attention retargeting, its connection to visual saliency, the challenges associated with it, and ideas for how it can be approached. The difficulty of attention retargeting as a saliency inversion problem lies in the lack of one-to-one mapping between saliency and the image domain, in addition to the possible negative impact of saliency alterations on image aesthetics. A few approaches from recent literature to solve this challenging problem are reviewed, and several suggestions for future development are presented.