Search

An image-map is a compromise between an image and a map. The quality of such maps is affected by several factors, such as (a) the matching between the features on images and the graphic symbols fro...

Image authentication based on image hashing has gained large attention in recent years. However, limited work has been done in color image hashing. Also, most of the existing methods are unable to detect tampering, if the composite rotation–scaling–translation (RST) distortion and tampering in a color image occur simultaneously. In this paper, an image hashing technique has been proposed based on convolutional stacked denoising auto-encoders (CSDAEs). In addition, a blind geometric correction approach is used to correct the composite RST distortion in the image. An input image is hierarchically mapped to a lower-dimensional hash code via CSDAEs, which have been trained for content-preserving operations (CPOs). An image map is generated from the hash via the decoder. The tampered area has been localized, by comparing the image map of hash codes from the reference image and the received image. The experimental results show that the proposed method is robust against most of the CPOs, especially to composite RST, a better trade-off between robustness and discrimination, and can localize the tampered regions. The receiver operating characteristics show that the proposed model is better than some of the state-of-the-art methods.

The microstructural evaluation of complex cementitious materials has been made possible by the microscopic imaging tools such as Scanning Electron Microscope (SEM) and X-Ray Microanalysis. Particularly, the application of concrete SEM imaging and digital image analysis have become common in the analysis and mapping of concrete technology. In this study, six samples of two-dimensional (2D) SEM images were spatially resampled to produce Geo-referenced SEM sample images. Subsequently, they were analyzed and the intensity histogram plot was produced to facilitate visual interpretation. The consecutive digital image analysis performed was the enhancement and noise removal process using two filtering methods i.e. median and adaptive box filter. The filtered resampled images, then undergone the unsupervised K-Means classification process to collectively separate each individual pixel corresponds to the spectral data. By spatial segmentation of K-Means algorithms, the cluster groups generated were carefully reviewed before proceeding to the final analysis. From the resulting data, the mapping of the spatial distribution of k-cluster and the quantification of micro-cracks (voids) were performed. The results of the SEM images (1st - 4th sample) showed a higher percentage of k-cluster data indicating a good correlation with the major elemental composition of EDX analysis, namely Oxide (O), Silicon (Si) and Carbon (C). Meanwhile, the subjective visual assessment of the image (5th and 6th sample) has confirmed the micro-crack developments on the concrete SEM images upon which the crack density was 3.02 % and 1.30 %, respectively.

High-resolution satellite images such as KOMPSAT-3 data provide detailed geospatial information over interest areas that are evenly located in an inaccessible area. The high-resolution satellite cameras are designed with a long focal length and a narrow field of view to increase spatial resolution. Thus, images show relatively narrow swath widths (10–15 km) compared to dozens or hundreds of kilometers in mid-/low-resolution satellite data. Therefore, users often face obstacles to orthorectify and mosaic a bundle of delivered images to create a complete image map. With a single mosaicked image at the sensor level delivered only with radiometric correction, users can process and manage simplified data more efficiently. Thus, we propose sensor-level mosaicking to generate a seamless image product with geometric accuracy to meet mapping requirements. Among adjacent image data with some overlaps, one image is the reference, whereas the others are projected using the sensor model information with shuttle radar topography mission. In the overlapped area, the geometric discrepancy between the data is modeled in spline along the image line based on image matching with outlier removals. The new sensor model information for the mosaicked image is generated by extending that of the reference image. Three strips of KOMPSAT-3 data were tested for the experiment. The data showed that irregular image discrepancies between the adjacent data were observed along the image line. This indicated that the proposed method successfully identified and removed these discrepancies. Additionally, sensor modeling information of the resulted mosaic could be improved by using the averaging effects of input data.

Image quality manipulating in JPEG is done by quantization tables. JPEG has two quantization tables – one table for the luminance information and one table for the chrominance information. These quantization tables have been designed in support of images with few sharp changes; however, typically most GPS image maps have many sharp changes and as a result, the images are not optimally compressed. The designers of the quantization tables have presumed that sharp changes in the colors will rarely occur. Therefore, they divide the values that represent sharp changes in the frequency space by large numbers and divide other values by smaller numbers. As a result, when there are sharp changes in an image, the proportional allocation for each kind of data in the compressed image is inappropriate and results in an inefficient compression. In this paper the standard quantization tables have been modified as to handle the different kinds of GPS image map data appropriately. Consequently, the experimental results show that images with sharp changes are compressed more efficiently when making use of the new quantization tables.

The brain–computer interface (BCI) control technology that utilizes motor imagery to perform the desired action instead of manual operation will be widely used in smart environments. However, most of the research lacks robust feature representation of multi-channel EEG series, resulting in low intention recognition accuracy. This article proposes an EEG2Image based Denoised-ConvNets (called EID) to enhance feature representation of the intention recognition task. Specifically, we perform signal decomposition, slicing, and image mapping to decrease the noise from the irrelevant frequency bands. After that, we construct the Denoised-ConvNets structure to learn the colorspace and spatial variations of image objects without cropping new training images precisely. Toward further utilizing the color and spatial transformation layers, the colorspace and colored area of image objects have been enhanced and enlarged, respectively. In the multi-classification scenario, extensive experiments on publicly available EEG datasets confirm that the proposed method has better performance than state-of-the-art methods.

With the progression of picture archiving and communication systems (PACSs) over the past decade, it has become imperative that such systems be optimised in security, storage, and transmission aspects. The work presented in this Letter shows a framework for medical image compression and secure image transmission for PACSs. The work aims to achieve a lower dimensionality of input medical image signified by a high-compression ratio, a secure image transmission that can withstand adversarial attacks and provide a reversible reconstruction with minimal error. The authors illustrate that sinusoid modulated Gaussian texture maps, multi-level chaotic maps, and high-frequency image maps can be efficiently fused and utilised in a deep learning architecture. The overall analysis depicts promising results with regard to the capability of image compression, security, and transmission. The proposed framework will be a potential candidate for use in PACSs, which effectively is the backbone of the current healthcare paradigm.

The semiconductor industry has become fully automated during the manufacturing process and abundant process parameters are collected online by sensors for fault detection and classification purposes. Analyzing process parameters and identifying a smaller set of key parameters that have crucial influence on wafer quality will bring great benefits in stabilizing the manufacturing process and enhancing productive yield. Typically, this type of the parameter set is called the “raw trace data.” This paper considers image processing techniques as a novel approach for analyzing and visualizing the raw trace data. First, the 1-D time series data of a wafer batch was transformed into a 2-D image. Fisher’s criterion ratios of the labeled good and defective wafer image maps are computed to identify the key parameters. The key parameters identified by the proposed image processing technique are consistent with the technical experience of the process engineers. Furthermore, the texture analysis technique with 2-D Fourier transform is utilized to analyze the images of the key parameters to detect defective wafers. The proposed key parameter identification and wafer classification method proves to be a viable solution under the paradigm of advanced process control practice for semiconductor manufacturing.

In applications of deep reinforcement learning to robotics, it is often the case that we want to learn pose invariant policies: policies that are invariant to changes in the position and orientation of objects in the world. For example, consider a pegin-hole insertion task. If the agent learns to insert a peg into one hole, we would like that policy to generalize to holes presented in different poses. Unfortunately, this is a challenge using conventional methods. This paper proposes a novel state and action abstraction that is invariant to pose shifts called deictic image maps that can be used with deep reinforcement learning. We provide broad conditions under which optimal abstract policies are optimal for the underlying system. Finally, we show that the method can help solve challenging robotic manipulation problems.

This paper presents a novel snapshot imaging spectrometer based on the image mapping and compressed sensing concept named Compressed Sensing Image Mapping Spectrometer (CSIMS). The operation principle is to slice the input image to different directions and encode the strip pieces before dispersion by a prism. The detector obtains the mixture spatial-spectral data simultaneously. The datacube is reconstructed by the compressed sensing algorithm and combining all the pieces together. The mathematical model of CSIMS is established to describe the light wave propagation through the entire system based on the scalar diffraction theory. The simulations are conducted to prove the effectiveness of the CSIMS principle, and the results show that the reconstructed datacube reveals higher spatial resolution and more accurate spectral curves than that of the relative snapshot imaging spectrometer based on compressed sensing.

Photography is the activity of recording precious moments which are often difficult to make up afterwards. Therefore, taking the correct picture under proper guidance assistance is important. Although there are many factors that can determine a good photo, in general, photos that do not follow the composition rules usually look bad that make the viewer feel uncomfortable. Acting as a solution, in this paper, we propose an intelligent photographing guidance system using machine learning. The guidance is based on a tree-based interpretable machine learning model that can give reasons for decisions. There are two categories of features for guidance, which are traditional image features and deep features. Traditional features include prominent lines and image maps, such as saliency map and sharpness map, each of which exists in a multi-scale Gaussian pyramid. Deep features are extracted during the establishment of a CNN-based image composition classifier. We use these two categories of features as inputs for the interpretable machine learning model to establish a feasible photographing guidance system. The guidance system references our composition classifier with precision rate of 94.8%, and recall rate of 95.0% where the comprising tree-based interpretable model is capable of guiding camera users to alter image contents for obtaining better aesthetical compositions to take photos of good quality.

Image filtering is a technique that can create additional visual representations of the original image. Entropy filtering is a specific application that can be used to highlight randomness of pixel grayscale intensities within an image. These image map created from filtering are based on the number of surrounding neighbourhood of pixels considered. However, there is no standard procedure for determining the correct "neighbourhood size" to use. We investigated the effects of neighbourhood size on the entropy calculation and provide a standardized approach for determining an appropriate neighbourhood size in entropy filtering in a musculoskeletal application. Ten healthy subjects showing no symptoms related to neuromuscular disease were recruited and ultrasound images of their trapezius muscle were acquired. The muscle regions in the images were manually isolated and regions of interest with varying neighbourhood sizes (increasing by 2 pixels) from 3x3 to 61X61 pixels were extracted. The entropy, relative signal entropy over noise entropy, statistical effect size as well as the percentage change of the effect size and instantaneous slope of the effect size was examined. The analysis showed that a neighbourhood size within the range of 21-25 pixels provides the maximum amount of information gained and coincides with a percentage change of the effect size of less than 5% and instantaneous slopes < 0.05.

Although existing image inpainting approaches have been able to produce visually realistic and semantically correct results, they produce only one result for each masked input. In order to produce multiple and diverse reasonable solutions, we present Unsupervised Cross-space Translation Generative Adversarial Network (called UCTGAN) which mainly consists of three network modules: conditional encoder module, manifold projection module and generation module. The manifold projection module and the generation module are combined to learn one-to-one image mapping between two spaces in an unsupervised way by projecting instance image space and conditional completion image space into common low-dimensional manifold space, which can greatly improve the diversity of the repaired samples. For understanding of global information, we also introduce a new cross semantic attention layer that exploits the long-range dependencies between the known parts and the completed parts, which can improve realism and appearance consistency of repaired samples. Extensive experiments on various datasets such as CelebA-HQ, Places2, Paris Street View and ImageNet clearly demonstrate that our method not only generates diverse inpainting solutions from the same image to be repaired, but also has high image quality.

Most existing dehazing methods are not robust to nonhomogeneous haze. Meanwhile, the information of dense haze region is usually unknown and hard to estimate, leading to blurry in dehaze result for those regions. Focusing on these two issues, we propose a novel coarse-to-fine model, namely Trident Dehazing Network (TDN), to learn the hazy to hazy- free image mapping with automatic haze density recognition. In detail, TDN is composed of three sub-nets: the Encoder-Decoder Net (EDN) is the main net of TDN to reconstruct the coarse hazy-free feature; the Detail Refinement sub-Net (DRN) helps to refine the high frequency details that was easily lost in the pooling layers in the encoder; and the Haze Density Map Generation sub-Net (HDMGN) can automatically distinguish the thick haze region with thin one, to prevent over-dehazing or under-dehazing in regions of different haze density. Moreover, we propose a frequency domain loss function to make supervision of different frequency band more uniform. Extensive experimental results on synthetic and real datasets demonstrate that our proposed TDN outperforms the state-of-the-arts with better fidelity and perceptual, generalizing well on both dense haze and nonhomogeneous haze scene. Our method won the first place in NTIRE2020 nonhomogeneous dehazing challenge.

We make an offer of a state-of-the-art method of the deep salient wood image-based quality assessment (DS-WIQA) for no-reference image appraisal. We explore a five-layer deep convolutional neural network (DCNN) for the salient wood image map. The DS-WIQA uses the concave n-square method. The outcomes allow that DS-WIQA model has a greater achievement on Zenodo and Lignoindo datasets, respectively. We appraise a salient wood image map by extracting in small wood image patches. The DS-WIQA has an admirable performance of other recent methods on Zenodo and Lignoindo datasets, respectively. DS-WIQA outdoes other recent techniques by 14.29% and 19.96% more advanced than other techniques with respect to SROCC and LCC measurement, respectively. DS-WIQA shows up to be more significant than the other DCNN methods.

Depression is a mental illness that may be harmful to an individual’s health. Using deep learning models to recognize the facial expressions of individuals captured in videos has shown promising results for automatic depression detection. Typically, depression levels are recognized using 2D-Convolutional Neural Networks (CNNs) that are trained to extract static features from video frames, which impairs the capture of dynamic spatio-temporal relations. As an alternative, 3D-CNNs may be employed to extract spatiotemporal features from short video clips, although the risk of overfitting increases due to the limited availability of labeled depression video data. To address these issues, we propose a novel temporal pooling method to capture and encode the spatio-temporal dynamic of video clips into an image map. This approach allows fine-tuning a pre-trained 2D CNN to model facial variations, and thereby improving the training process and model accuracy. Our proposed method is based on two-stream model that performs late fusion of appearance and dynamic information. Extensive experiments on two benchmark AVEC datasets indicate that the proposed method is efficient and outperforms the state-of-the-art schemes.

In recent years, an evolutionary change occurred in the digital world when various social networking sites became popular. This led to a big increase of users who share their activities in digital form. A huge amount of digital information has become available, providing researchers with a unique insight into the behavior and activities of entire population. The use of Geo-tagged social media data became an emerging trend to represent user actions and behaviour at specific geographical locations. However, the inaccuracies of Geo-tagged data from social media often limits the utility of this data source in micro scale analysis (at the street or place of interest level (POI)). Our study supports how social media data could be matched accurately with specific physical locations. More specifically, our investigation includes Geo-tweet data and image data from flickr to map accurately physical location using machine leaning and deep learning techniques. In this paper, a preliminary discussion of this work is provided, using Geo-tagged data from Twitter and Open StreetMap in cities such as San Francisco and London as well as for image mapping using Flickr data in cities such as San Francisco and Kyoto.

In this paper, we present an image classification approach to action recognition with 3D skeleton videos. First, we propose a video domain translation-scale invariant image mapping, which transforms the 3D skeleton videos to color images, namely skeleton images. Second, a multi-scale dilated convolutional neural network (CNN) is designed for the classification of the skeleton images. Our multi-scale dilated CNN model could effectively improve the frequency adaptiveness and exploit the discriminative temporal-spatial cues for the skeleton images. Even though the skeleton images are very different from natural images, we show that the fine-tuning strategy still works well. Furthermore, we propose different kinds of data augmentation strategies to improve the generalization and robustness of our method. Experimental results on popular benchmark datasets such as NTU RGB + D, UTD-MHAD, MSRC-12 and G3D demonstrate the superiority of our approach, which outperforms the state-of-the-art methods by a large margin.

This paper presents a fast blind image sharpness/blurriness assessment model (BISHARP) which operates in spatial and transform domain. The proposed model generates local contrast image maps by computing the root-mean-squared values for each image pixel within a defined size of local neighborhood. The resulting local contrast maps are then transformed into the wavelet domain where the reduction of high frequency content is evaluated in the presence of varying blur strengths. It was found that percentile values computed from sorted, level-shifted, high-frequency wavelet coefficients can serve as reliable image sharpness/blurriness estimators. Furthermore, it was found that higher dynamic range of contrast maps significantly improves model performance. The results of validation performed on seven image databases showed a very high correlation with perceptual scores. Due to low computational requirements the proposed model can be easily utilized in real-world image processing applications.

This paper aims to develop a vision-based driver assistance system for scene awareness using video frames obtained from a dashboard camera. A saliency image map is devised with features pertinent to the driving scene. This saliency map mimics the human contour and motion sensitive visual perception by extracting spatial, spectral, and temporal information from the input frames and applying entropy driven image-context-feature data fusion. The resultant fusion output comprises high-level descriptors for still segment boundaries and non-stationary object appearance. Following the segmentation and foreground object detection stage, an adaptive maximum likelihood classifier selects road surface regions. The proposed scene driven vision system improves the driver’s situational awareness by enabling adaptive road surface classification. As experimental results demonstrate, context-aware low-level to high-level information fusion based on human vision model produces superior segmentation, tracking, and classification results that lead to high- level abstraction of driving scene.

Image conversion has attracted mounting attention due to its practical applications. This paper proposes a lightweight network structure that can implement unpaired training sets to complete one-way image mapping, based on the generative adversarial network (GAN) and a fixed-parameter edge detection convolution kernel. Compared with the cycle consistent adversarial network (CycleGAN), the proposed network features simpler structure, fewer parameters (only 37.48% of the parameters in CycleGAN), and less training cost (only 35.47% of the GPU memory usage and 17.67% of the single iteration time in CycleGAN). Remarkably, the cyclic consistency becomes not mandatory for ensuring the consistency of the content before and after image mapping. This network has achieved significant processing effects in some image translation tasks, and its effectiveness and validity have been well demonstrated through typical experiments. In the quantitative classification evaluation based on VGG-16, the algorithm proposed in this paper has achieved superior performance.

Auto-tuning and parametric implementation of deep learning kernels allow off-the-shelf accelerator-based embedded platforms to deliver high-performance and energy-efficient mappings of the inference phase of lightweight neural networks. Low-complexity classifiers are characterized by operations on small image maps with two to three deep layers and few class labels. For these use cases, we consider a range of embedded systems with 20W power budgets such as the Xilinx ZC706 (FPGA), NVIDIA Jetson TX1 (GPU), TI Keystone II (DSP), and Adapteva Parallella (RISC+NoC). In CaffePresso, we combine auto-tuning of the implementation parameters, and platform-specific constraints deliver optimized solutions for each input ConvNet specification.

In this paper, a novel framework, named as global-local feature attention network with reranking strategy (GLAN-RS), is presented for image captioning task. Rather than only adopting unitary visual information in the classical models, GLAN-RS explores the attention mechanism to capture local convolutional salient image maps. Furthermore, we adopt reranking strategy to adjust the priority of the candidate captions and select the best one. The proposed model is verified using the Microsoft Common Objects in Context (MSCOCO) benchmark dataset across seven standard evaluation metrics. Experimental results show that GLAN-RS significantly outperforms the state-of-the-art approaches, such as multimodal recurrent neural network (MRNN) and Google NIC, which gets an improvement of 20% in terms of BLEU4 score and 13 points in terms of CIDER score.

The problem of overcrowding of mega cities has been bolded in recent years. To meet the need of housing this increased population, which is of great importance in mega cities, a huge number of buildings are constructed annually. With the ever-increasing trend of building constructions, we are faced with the growing trend of building infractions and illegal buildings (IBs). Acquiring multi-temporal satellite images and using change detection techniques is one of the proper methods of IB monitoring. Using the type of satellite images with different spatial and spectral resolutions has always been an issue in efficient detection of the building changes. In this research, three bi-temporal high-resolution satellite images of IRS-P5, GeoEye-1 and QuickBird sensors acquired from the west of metropolitan area of Tehran, capital of Iran, in addition to city maps and municipality property database were used to detect the under construction buildings with improved performance and accuracy. Furthermore, determining the employed bi-temporal satellite images to provide better performance and accuracy in the case of IB detection is the other purpose of this research. The Kappa coefficients of 70 %, 64 %, and 68 % were obtained for producing change image maps using GeoEye-1, IRS-P5, and QuickBird satellite images, respectively. In addition, the overall accuracies of 100 %, 6 %, and 83 % were achieved for IB detection using the satellite images, respectively. These accuracies substantiate the fact that the GeoEye-1 satellite images had the best performance among the employed images in producing change image map and detecting the IBs.

Monitoring of impact and multi-impact in particular in aircraft composite structures has been an intensive research topic in the field of guided-wave-based structural health monitoring (SHM). Compared with the majority of existing methods such as those using signal features in the time-, frequency- or joint time-frequency domain, the approach based on spatial-wavenumber filter of guided wave shows superb advantage in effectively distinguishing particular wave modes and identifying their propagation direction relative to the sensor array. However, there exist two major issues when conducting online characterization of multi-impact event. Firstly, the spatial-wavenumber filter should be realized in the situation that the wavenumber of high spatial resolution of the complicated multi-impact signal cannot be measured or modeled. Secondly, it’s difficult to identify the multiple impacts and realize multi-impact localization due to the overlapping of wavenumbers. To address these issues, a scanning spatial-wavenumber filter based diagnostic imaging method for online characterization of multi-impact event is proposed to conduct multi-impact imaging and localization in this paper. The principle of the scanning filter for multi-impact is developed first to conduct spatial-wavenumber filtering and to achieve wavenumber-time imaging of the multiple impacts. Then, a feature identification method of multi-impact based on eigenvalue decomposition and wavenumber searching is presented to estimate the number of impacts and calculate the wavenumber of the multi-impact signal, and an image mapping method is proposed as well to convert the wavenumber-time image to an angle-distance image to distinguish and locate the multiple impacts. A series of multi-impact events are applied to a carbon fiber laminate plate to validate the proposed methods. The validation results show that the localization of the multiple impacts are well achieved.

Unmanned Aerial Vehicles (UAVs) can be used to acquire highly accurate data in deformation survey, whereby low-cost digital cameras are commonly used in the UAV mapping. Thus, camera calibration is considered important in obtaining high-accuracy UAV mapping using low-cost digital cameras. The main focus of this study was to calibrate the UAV camera at different camera distances and check the measurement accuracy. The scope of this study included camera calibration in the laboratory and on the field, and the UAV image mapping accuracy assessment used calibration parameters of different camera distances. The camera distances used for the image calibration acquisition and mapping accuracy assessment were 1.5 metres in the laboratory, and 15 and 25 metres on the field using a Sony NEX6 digital camera. A large calibration field and a portable calibration frame were used as the tools for the camera calibration and for checking the accuracy of the measurement at different camera distances. Bundle adjustment concept was applied in Australis software to perform the camera calibration and accuracy assessment. The results showed that the camera distance at 25 metres is the optimum object distance as this is the best accuracy obtained from the laboratory as well as outdoor mapping. In conclusion, the camera calibration at several camera distances should be applied to acquire better accuracy in mapping and the best camera parameter for the UAV image mapping should be selected for highly accurate mapping measurement.

This paper reviews more than 10 recent doctoral theses and master’s theses accomplished at the Chair of Cartography, Technical University of Munich (TUM). These research works address two persistent challenges involved in the digital mapmaking process: understanding complex input data and improving cartographic design to keep up with the changing roles and increasing demands of users on immediately usable maps. At first, a dedicated computational approach based on Conditional Random Fields (CRF) for data enhancement of Volunteered Geographic Information (VGI) is introduced and demonstrated with test data from trajectories of floating taxis and OpenStreetMap (OSM). It is followed by a summarized overview of visual analytical approaches for event and behavior discovery and their implementations on various VGI data sources including trajectories of floating taxis, twitter messages and trajectories of football matches. Finally, a concept of image map with a multilayered visual hierarchy is proposed and demonstrated with a set of attention-guided design strategies. Keywords: Probabilistic Graphical Model, visual analytics, event and behavior discovery, attention-guided design Der vorliegende Beitrag gibt einen Uberblick uber die Forschungsinhalte von mehr als 10 aktuellen Dissertationen und Masterarbeiten, die am Lehrstuhl fur Kartographie der Technischen Universitat Munchen durchgefuhrt wurden. Diese Arbeiten befassen sich mit zwei dauerhaften Herausforderun- gen in der digitalen Kartographie: Verstehen von komplexen Eingabedaten und Verbesserung der Gestaltungsverfahren, um mit den sich verandernden Rollen und den stetig wachsenden Anforderungen der Kartennutzer mitzuhalten. Zunachst wird ein auf Probabilistischen Graphischen Modellen (PGM) basierter Ansatz zur Datenanreicherung der VGI vorgestellt und anhand Taxifahrtrouten und Strasennetz von OpenStreetMap (OSM) demonstriert. Es folgt eine Darstellung von visuellen analy- tischen Ansatzen sowie deren Umsetzung zur Entdeckung des Events und des Verhaltens von Personen in verschiedenen VGI-Datenquellen wie z.B. Trajektorien der Floating Taxis, Twitternachrichten und Trajektorien des Fusballspiels. Schlieslich wird das Konzept der Bildkarte, die in der Regel aus Kartensymbolen im Vordergrund und einem Rasterbild im Hintergrund besteht, in Frage gestellt. Eine Reihe von Aufmerk- samkeitsgetriebenen Gestaltungsstrategien zeigen, wie sich Bildkarten mit mehreren visuellen Ebenen generieren lassen. Schlusselworter: Probabilistische Graphische Modelle, Visuelle Analytik, Entdeckung des Events und Verhaltens, Aufmerksamkeits-betriebene Gestaltung

Thanks to the development of mobile mapping technologies, close-range photogrammetry ( crp ) has advanced to be an efficient mapping method for a variety of applications. A compact crp system equipped with multiple cameras and a gps receiver is one of those advanced portable mapping systems. A portable panoramic image mapping system ( ppms ) was specially designed to capture panoramic images with eight cameras and to obtain the position of image station with a gps receiver. A ppims can be considered as a panoramic crp system. The coordinates of an object point can be determined by the intersection of panoramic image points. For the implementation, we propose a new concept of photogrammetry by using panoramic images. Eight images captured by ppims forms a spherical panorama image ( spi ). Instead of using the original images, ppims spi s are then used for photogrammetric triangulation and mapping. Under this circumstance, one spi is formed for each station, and it is associated with only one set of exterior orientation ( eo ) parameters. Traditional collinearity equations are not applicable to spi triangulation and mapping. Therefore, a novel bundle adjustment algorithm is proposed to solve eo of multi-station spi s. Because ppims spi s are not ideal spi s, a correction scheme was also developed to correct the imperfect geometry of ppims spi . Two test studies were performed for the data collected at a campus test field of National Cheng Kung University ( ncku ) and at a historical site of Tainan. Both cases demonstrate the feasibility of spi bundle adjustment and applying corrections for ppims spi s necessary for effective for bundle adjustment. Furthermore, the experiment's results also confirm that spi s can replace original images for ppims triangulation.

Development of mapping and remote sensing to detection of green areas in a wide range can do aerial photography using drones. The aerial photo in question is a small format aerial photo using a camera. The image produced from aerial photographs is still fragmented into separate parts. Therefore, it is necessary to merge each sequential image. Merging is done by detecting the mapping of the area by sewing each image based on the point of similarity in pixels. The method applied with the search for similar features uses the Speeded Up Robust Features (SURF). The results obtained to see the level of similarity in the feature mapping area so that the merger into one detected area does not require a long time. The SURF method is applied, giving the results of the number of images that correspond to the Minimum Mean Square Error (MSE) level of 0.0246. The results obtained are the level of similarity at matched point 32 gives a panoramic view approaching the mapping according to the green area of the aerial photo.

Background The protocol in this manuscript was designed to help graduate students publish. It is the result of a challenge from our provost in 2013. I developed this protocol over the last 6 years and have exercised the protocol for the last 5 years. The current version of the protocol has remained mostly static for the last 2 years—only small changes have been made to the process. Objective The objective of this protocol is to enable students to learn a valuable skill of conducting a systematic review and to write the review in a way that can be published. I have designed the protocol to fit into the schedule of a traditional semester, but also used it in compressed semesters. Methods An image map was created in HTML 5.0 and imported into a learning management system. It augments traditional instruction by providing references to published articles, examples, and previously recorded instructional videos. Students use the image map outside the classroom after traditional instruction. The image map helps students create manuscripts that follow established practice and are reported in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA), and whose authorship follows guidelines by the International Committee of Medical Journal Editors. Results Since its inception, this protocol has helped 77 students publish 27 systematic reviews in nine journals worldwide. Some manuscripts take multiple years to progress through multiple review processes at multiple journals submitted in sequence. Two other professors in the School of Health Administration have used this protocol in their classes. Conclusions So far, this method has helped 51% of graduate students who used it in my graduate courses publish articles (with more manuscripts under consideration whose numbers have remained uncounted in this sum). I wish success to others who might use this protocol.

To achieve the robust high-performance computer-aided diagnosis systems for lymph nodes, CT images may be typically collected from multicenter data, which cause the isolated performance of the model based on different data source centers. The variability adaptation problem of lymph node data which is related to the problem of domain adaptation in deep learning differs from the general domain adaptation problem because of the typically larger CT image size and more complex data distributions. Therefore, domain adaptation for this problem needs to consider the shared feature representation and even the conditioning information of each domain so that the adaptation network can capture significant discriminative representations in a domain-invariant space. This paper extracts domain-invariant features based on a cross-domain confounding representation and proposes a cycle-consistency learning framework to encourage the network to preserve class-conditioning information through cross-domain image translations. Compared with the performance of different domain adaptation methods, the accurate rate of our method achieves at least 4.4% points higher under multicenter lymph node data. The pixel-level cross-domain image mapping and the semantic-level cycle consistency provided a stable confounding representation with class-conditioning information to achieve effective domain adaptation under complex feature distribution.

It is impossible to inspect or get an overview of image collections with millions of images. Users often start “exploring” images with a keyword or a similarity search. Both lead to long unstructured lists of result images. In this demo we present a graph-based system for visually exploring and navigating continuously changing sets of millions of images with a web browser. Subsets of images are successively retrieved from a image similarity graph and displayed as a visually sorted 2D image map, which can be zoomed and dragged to explore images from related concepts.

The growing use of virtual autonomous agents in applications like games and entertainment demands better control policies for natural-looking movements and actions. Unlike the conventional approach of hard-coding motion routines, we propose a deep learning method for obtaining control policies by directly mimicking raw video demonstrations. Previous methods in this domain rely on extracting low-dimensional features from expert videos followed by a separate hand-crafted reward estimation step. We propose an imitation learning framework that reduces the dependence on hand-engineered reward functions by jointly learning the feature extraction and reward estimation steps using Generative Adversarial Networks (GANs). Our main contribution in this paper is to show that under injective mapping between low-level joint state (angles and velocities) trajectories and corresponding raw video stream, performing adversarial imitation learning on video demonstrations is equivalent to learning from the state trajectories. Experimental results show that the proposed adversarial learning method from raw videos produces a similar performance to state-of-the-art imitation learning techniques while frequently outperforming existing hand-crafted video imitation methods. Furthermore, we show that our method can learn action policies by imitating video demonstrations on YouTube with similar performance to learned agents from true reward signals. Please see the supplementary video submission at https://ibm.biz/BdzzNA., Updated the paper to match with version accepted at IEEE MMSP 2019

With the rapid development of Chinese economy, the pace of urban construction has been accelerated obviously. At the same time, illegal land use and illegal construction are becoming more and more serious. At present, using satellite remote sensing technology to monitor land use dynamics is the main task of Law Enforcement Inspection of Satellite Land Image. 3S technology plays an important role in satellite film law enforcement and inspection. High-definition image maps acquired by RS in different periods are the basis of satellite film law enforcement. GPS can accurately locate dynamic patches. GIS technology can be used to construct basic geographic information database and spatial analysis of data. Therefore, taking 3S as the technical core, building a land digital monitoring platform greatly improves the efficiency of satellite film law enforcement and inspection.

Homogeneous image change detection research has been well developed, and many methods have been proposed. However, change detection between heterogeneous images is challenging since heterogeneous images are in different domains. Therefore, direct heterogeneous image comparison in the way that we do it is difficult. In this paper, a method for heterogeneous synthetic aperture radar (SAR) image and optical image change detection is proposed, which is based on a pixel-level mapping method and a capsule network with a deep structure. The mapping method proposed transforms an image from one feature space to another feature space. Then, the images can be compared directly in a similarly transformed space. In the mapping process, some image blocks in unchanged areas are selected, and these blocks are only a small part of the image. Then, the weighted parameters are acquired by calculating the Euclidean distances between the pixel to be transformed and the pixels in these blocks. The Euclidean distance calculated according to the weighted coordinates is taken as the pixel gray value in another feature space. The other image is transformed in a similar manner. In the transformed feature space, these images are compared, and the fusion of the two different images is achieved. The two experimental images are input to a capsule network, which has a deep structure. The image fusion result is taken as the training labels. The training samples are selected according to the ratio of the center pixel label and its neighboring pixels’ labels. The capsule network can improve the detection result and suppress noise. Experiments on remote sensing datasets show the final detection results, and the proposed method obtains a satisfactory performance.

The use of hyperspectral imaging techniques in biological studies has increased in the recent years. Hyperspectral images (HSI) provide chemical information and preserve the morphology and original structure of heterogeneous biological samples, which can be potentially useful in environmental –omics studies when effects due to several factors, e.g., contaminant exposure, phenotype,… at a specific tissue level need to be investigated. Yet, no available strategies exist to exploit adequately this kind of information. This work offers a novel chemometric strategy to pass from the raw image information to useful knowledge in terms of statistical assessment of the multifactor effects of interest in –omic studies. To do so, unmixing of the hyperspectral image measurement is carried out to provide tissue-specific information. Afterwards, several specific ANOVA-Simultaneous Component Analysis (ASCA) models are generated to properly assess and interpret the diverse effect of the factors of interest on the spectral fingerprints of the different tissues characterized. The unmixing step is performed by Multivariate Curve Resolution-Alternating Least Squares (MCR-ALS) on multisets of biological images related to each studied condition and provides reliable HSI spectral signatures and related image maps for each specific tissue in the regions imaged. The variability associated with these signatures within a population is obtained through an MCR-based resampling step on representative pixel subsets of the images analyzed. All spectral fingerprints obtained for a particular tissue in the different conditions studied are used to obtain the related ASCA model that will help to assess the significance of the factors studied on the tissue and, if relevant, to describe the associated fingerprint modifications. The potential of the approach is assessed in a real case of study linked to the investigation of the effect of exposure time to chlorpyrifos-oxon (CPO) on ocular tissues of different phenotypes of zebrafish larvae from Raman HSI of eye cryosections. The study allowed the characterization of melanin, crystalline and internal eye tissue and the phenotype, exposure time and the interaction of the two factors were found to be significant in the changes found in all kind of tissues. Factor-related changes in the spectral fingerprint were described and interpreted per each kind of tissue characterized. © 2018 Elsevier B.V., The research leading to these results has received funding from the European Research Council under the European Union's Seventh Framework Programme ( FP/2007-2013 ) / ERC Grant Agreement n. 32073 (CHEMAGEB project). The authors of this work belong to the network of recognized research groups by the Catalan government (2017 SGR 753 / 2017 SGR 902) and acknowledge the support of the Spanish government through project CTQ2015-66254-C2-2-P , and from NATO (SfP project MD.SFPP 984777 ). ICFO acknowledges financial support from the Spanish Ministry of Economy and Competitiveness through the “Severo Ochoa” program for Centres of Excellence in R&D ( SEV-2015-0522 ), from Fundación Cellex , Fundación Mig-Puig and from Generalitat de Catalunya through the CERCA program", Laserlab-Europe ( EU-H2020 654148 ), Fundació la Marató de TV3 (Molecular imaging of the retina in patients with Multiple Sclerosis by Raman Spectroscopy, 20142030 ), and the National Institutes of Health (NIH, grant 5R21CA187890-02 ). Appendix A

In Additive Manufacturing AM, each volume of material experiences a complex thermal history due to both short-range effects, from the repeated overlap of the thermal field from each heat source pass, and long-range variation in the thermal boundary conditions, related to the part geometry and build height. With an α + β alloy, like Ti64, this can lead to significant local variation in the transformation microstructure, which can contribute to heterogeneity in the mechanical properties of a component. In order to better understand the transformation microstructure variability in AM parts, an automated microstructure analysis tool has been developed, and tested against independently measured data, that can accurately map the inter-lamellar spacing of the α phase and spheroidicity of the β phase, at both high resolution and over large distances. The approach used was based on automated batch image analysis of thousands of image tiles obtained using a mapping function in a high-resolution SEM with a scanning stage. Within a practical operating range of drift in the microscope parameters (e.g. working distance, detector contrast) the errors in the measurements were found to be minimal (

We present a novel approach based on Residual Generative Adversarial Network (R-GAN) to discriminate genuine pain expression from posed pain expression by magnifying the subtle changes in the face. In addition to the adversarial task, the discriminator network in R-GAN estimates the intensity level of the pain. Moreover, we propose a novel Weighted Spatiotemporal Pooling (WSP) to capture and encode the appearance and dynamic of a given video sequence into an image map. In this way, we are able to transform any video into an image map embedding subtle variations in the facial appearance and dynamics. This allows using any pre-trained model on still images for video analysis. Our extensive experiments show that our proposed framework achieves promising results compared to state-of-the-art approaches on three benchmark databases, i.e., UNBC-McMaster Shoulder Pain, BioVid Head Pain, and STOIC.

Incorporating human domain knowledge for breast tumor diagnosis is challenging, since shape, boundary, curvature, intensity, or other common medical priors vary significantly across patients and cannot be employed. This work proposes a new approach for integrating visual saliency into a deep learning model for breast tumor segmentation in ultrasound images. Visual saliency refers to image maps containing regions that are more likely to attract radiologists visual attention. The proposed approach introduces attention blocks into a U-Net architecture, and learns feature representations that prioritize spatial regions with high saliency levels. The validation results demonstrate increased accuracy for tumor segmentation relative to models without salient attention layers. The approach achieved a Dice similarity coefficient of 90.5 percent on a dataset of 510 images. The salient attention model has potential to enhance accuracy and robustness in processing medical images of other organs, by providing a means to incorporate task-specific knowledge into deep learning architectures., Comment: 16 pages, 5 figures

Optimizing a network of maps among a collection of objects/domains (or map synchronization) is a central problem across computer vision and many other relevant fields. Compared to optimizing pairwise maps in isolation, the benefit of map synchronization is that there are natural constraints among a map network that can improve the quality of individual maps. While such self-supervision constraints are well-understood for undirected map networks (e.g., the cycle-consistency constraint), they are under-explored for directed map networks, which naturally arise when maps are given by parametric maps (e.g., a feed-forward neural network). In this paper, we study a natural self-supervision constraint for directed map networks called path-invariance, which enforces that composite maps along different paths between a fixed pair of source and target domains are identical. We introduce path-invariance bases for efficient encoding of the path-invariance constraint and present an algorithm that outputs a path-variance basis with polynomial time and space complexities. We demonstrate the effectiveness of our approach on optimizing object correspondences, estimating dense image maps via neural networks, and semantic segmentation of 3D scenes via map networks of diverse 3D representations. In particular, for 3D semantic segmentation, our approach only requires 8% labeled data from ScanNet to achieve the same performance as training a single 3D segmentation network with 30% to 100% labeled data.

The underwater environment makes object-detection missions difficult. Side-scan Sonar (SSS) has been found to be suitable for seabed scanning missions, however the sonar images acquired from SSS often suffer from considerable noise and geometrical distortion, which changes the understanding of the texture, size, and shape of seabed objects. In order to identify seabed objects, it is thus vital to reconstruct the actual shape by reducing distortion. This paper proposes a process for correcting and reconstructing the sonar image map that utilizes intensity normalization, slant range correction, yaw and pitch correction, and speed and location correction. This is done using navigation and inertial data acquired by the autonomous underwater vehicle sensors.

Fault diagnosis is important to ensure the safety and efficience of mechanical equipment. In recent years, data-driven fault diagnosis methods have received extensive attention and research from many scholars. Different from the traditional fault feature extraction methods that rely on expert experience, this paper proposes a feature extraction method based on deep learning (DL). In order to meet the needs of neural networks for the amount of data, data augmentation is emploied to increase the amount of original data. Then, a novel signal-to-image mapping (STIM) is proposed to convert the one-dimensional vibration signals into two-dimensional grey images, which greatly reduce the human involvement. Finally, a convolutional neural network (CNN) model is established to extract fault features from grey images and realize fault classification. The learning process of the CNN model is analyzed and two different bearing experiment datasets are used to verify the effectiveness of the proposed method.

Field Emission devices (FE) have been proposed as efficient electron sources for several applications such as electron microscopy and vacuum sensors. Evidently, characterization methods applied during development phase of FE devices are crucial to evaluate aspects related with their working stability, homogeneity, and efficiency. However, the traditional methods provide only overall information about such characteristics, which difficult to improve the performance of these devices and their integration with electronics. To overcome this problem, this work presents an alternative system to characterize FE devices through electron emission imaging in real-time. The proposed system acquires I - V features of FE devices, while a video camera captures the emission image from a phosphor screen. Virtual instrumentation based on LabVIEW manages the whole system including measurement instruments, image capture, and data processing. As a result, histograms, 3D maps, and other FE analyses provide information about emitting characteristics of selected regions of interest. The main contribution of this work is to offer an important tool for the analyses of electron emission, by the association of captured images with the localized emission current. The extracted information from our system can efficiently support the characterization and the development of FE devices.

Image registration (IR) aims to map one image to another of a same scene. With rapid progress in image acquisition technologies, 3D IR becomes an important problem in magnetic resonance imaging (MRI) and other applications. In the literature, however, most IR methods are for 2D images and there are only a limited number of 3D methods available. Because 3D images have much complicated structure than their 2D counterparts, 3D IR is not just a simple generalization of the 2D IR problem. In this paper, we develop a 3D IR method that can handle cases with affine geometric transformations well. By its definition, an affine transformation maps a line to a line, and it includes rotation, translation, and scaling as special cases. In practice, most geometric transformations involved in IR problems are affine transformations. Therefore, our proposed method can find many IR applications. It is shown that this method works well in various cases, including cases when the data size of a 3D image is reduced for different reasons. This latter property makes it attractive for many 3D IR applications, since 3D images are often big in data size and it is natural to reduce their size for fast computation.

Small tunnels such as engineering geological exploratory tunnels and mine roadways are generally narrow, which make the existing photographic geological logging technique inapplicable. Therefore, geological logging of exploratory tunnels has always been taking the method of manual sketch work which has low efficiency and poor informatization degree of products, and it is a technical issue requiring urgent settlement for geological logging of small tunnels. This paper proposes and studies novel methods of photographic geological logging suitable for small tunnels, including image acquisition, image orientation control, image geometric correction, unfolded image map generation and geological attitude measurement, etc. Experiments show that the method can meet the precision requirement of geological logging. The novel method helps to realize the fast acquisition and processing of image-based geological logging data for small tunnels, and the forms of logging result are more abundant and more applicable to informatized management and application of geological logging data.

The quality of elemental image maps obtained via line scan-based LA-ICPMS is a function of the temporal response of the entire system, governed by the design of the system and mapping and acquisition conditions used, next to the characteristics of the sample. To quantify image degradation, ablation targets with periodic gratings are required for the construction of a modulation transfer function (MTF) and subsequent determination of the lateral resolution as a function of image noise and contrast. Since such ablation targets, with suitable matrix composition, are not readily available, computer-generated periodic gratings were virtually ablated via a computational process based on a two-step discrete-time convolution procedure using empirical/experimental input data. This experimental-modeling procedure simulates LA-ICPMS imaging based on two consecutive processes, viz., LA sampling (via ablation crater profiles [ACP]) and aerosol washout/transfer/ICPMS measurement (via single pulse responses [SPR]). By random selection of experimental SPRs from a large database for each individual pulse during the simulation, the convolution procedure simulates an accurate elemental image map of the periodic gratings with realistic (proportional or flicker) noise. This facilitates indirect retrieval of the experimental lateral resolution for the matrix targeted without performing actual line scanning on periodic gratings.

This paper presents an underwater structure 3D reconstruction method using a 2D multibeam imaging sonar. Compared with other underwater environmental recognition sensors, the 2D multibeam imaging sonar offers high resolution images in water with a high turbidity level by showing the reflection intensity data in real-time. With such advantages, almost all underwater applications, including ROVs, have applied this 2D multibeam imaging sonar. However, the elevation data are missing in sonar images, which causes difficulties with correctly understanding the underwater topography. To solve this problem, this paper concentrates on the physical relationship between the sonar image and the scene topography to find the elevation information. First, the modeling of the sonar reflection intensity data is studied using the distances and angles of the sonar beams and underwater objects. Second, the elevation data are determined based on parameters like the reflection intensity and shadow length. Then, the elevation information is applied to the 3D underwater reconstruction. This paper evaluates the presented real-time 3D reconstruction method using real underwater environments. Experimental results are shown to appraise the performance of the method. Additionally, with the utilization of ROVs, the contour and texture image mapping results from the obtained 3D reconstruction results are presented as applications.

Image maps have become very popular and frequently produced cartographical outputs during recent years. However, the unambiguous terminology, definitions, content and appearance specification have not been widely researched. The paper deals with the new definition of image map, its components delineation, and basic classification. The authors understand the image map as a special map portraying geographic space in a particular cartographical projection and map scale, where its content consists of two basic components – image and symbol components. Image component is represented by remote sensing image(s), while symbol component is represented by cartographical symbols. An image map has to have three essential attributes: cartographical projection, map scale and symbol component by means of map language. The authors also present aspects of topographic and thematic image maps.