Masked Area Research Articles

Structure-guided Generative Adversarial Network for Image Inpainting

Abstract Generative Adversarial Network based image inpainting algorithms often make errors when filling arbitrary masked areas because all input pixels are treated as effective pixels during convolutional operations. To resolve this matter, we present a novel solution: an image inpainting algorithm that utilizes gated convolutions within the residual blocks of the network. By incorporating gated convolutions instead of traditional convolutions, our algorithm effectively learns and captures the relationship between the known regions and the masked regions. The algorithm utilizes a two-stage generative adversarial restoration network, where the structure and texture restoration are performed sequentially. Specifically, the structural information of the known region in the damaged image is detected using an edge detection algorithm. Subsequently, the edges of the masked area are combined with the color and texture information of the known region for structure restoration. Finally, the complete structure and the image to be restored are fed into the texture restoration network for texture restoration, yielding the complete image output. During network training, a spectral normalization Markovian discriminator is employed to address the slow weight changes during iteration, thereby increasing convergence speed and model accuracy. Based on the Places2 dataset, our experimental findings indicate that our algorithm surpasses existing two-stage restoration algorithms in terms of improving peak signal-to-noise ratio and structural similarity. Specifically, our proposed algorithm achieves a 4.3% enhancement in peak signal-to-noise ratio and a 3.7% improvement in structural similarity when restoring images with various shapes and sizes of damaged areas. Additionally, it produces noticeable visual enhancements, further validating its effectiveness.

Unveiling human eye temperature with deep learning-powered segmentation

Thermal imaging technology presents a promising, non-invasive tool for medical diagnostics, particularly in eye health assessment. However, accurately delineating the eye region in thermal images of the human face remains a significant challenge for temperature analysis. Extracting eye temperature information from ocular thermal images has traditionally been a manual or semi-automated process, making it time-consuming and tedious. Our research aims to address these issues by introducing an automated systematic approach to eye segmentation and temperature analysis. We employ the semantic segmentation deep learning model SegFormer to segment the eye region and a novel method to extract the thermal-ocular temperature within this region of interest (ROI) for analysis of temperature patterns and diagnosis of any abnormalities. The process starts by extracting temperature data from each pixel in the thermal images. Using the mask and these temperature values, only the temperatures within the masked area are isolated for analysis. Validation is done by comparing and calculating the mean square error against the manual method. A self-created dataset of thermal images of human faces is used due to the lack of available datasets for segmentation. SegFormer (B5) achieves a mean intersection over union (MIoU) of 0.9178 and a dice coefficient of 0.9299. The extracted temperature values from the segmented eye regions provide crucial data for assessing eye health and identifying potential abnormalities. This methodical approach contributes to the progression of thermal image analysis for diagnosing eye-related conditions, particularly in ophthalmology, and in medical domains such as fever screening and reliable Computer-Aided Diagnosis Systems (CADS).

MDTNet: Partial transformer with degradation-aware module for restoring old photos with multiple degradations

Old photos preserve the memories of the old days and have special meanings in life. However, they often suffer from severe and complex degradations, such as cracks, dirt, and noise. Old photo restoration is thus a meaningful but challenging task. Most of the current deep learning methods to repair corrupted old photos focus on one single type of degradation or limited number of degradations. The lack of proper old photo datasets makes the digital restoration problem even more challenging. In this paper, we propose MDTNet, which is a novel transformer-based method for restoring old photos with multiple degradations. MDTNet is an efficient end-to-end transformer architecture with only a single encoder and a decoder to complete the restoration. Specifically, a novel partial transformer encoder is designed with partial multi-head self-attention to extract local-global context information from the valid masked area, which represents the unbroken area. In addition, a degradation-aware module is proposed in the decoder to automatically learn different degradation information, to improve the effect of multiple degradation restoration. Moreover, a synthetic old photo dataset — the SynOld dataset — was built, which simulates the multiple degradations of real old photos. Experimental results show that the proposed method outperforms the state-of-the-art methods on both public datasets and the custom-developed dataset. The code is available at https://github.com/YuanZhaoc/MDTNet.

Mask wearing impacts skin barrier function and microbiome profile in sensitive skin

Mask-wearing behavior, common in the post-COVID-19 era, raises concerns for sensitive skin. This split-face study investigated mask-related changes in skin barrier function and microbiome composition among 30 female volunteers with sensitive skin and assessed the mitigating effects of a moisturizer containing biological lipids and probiotics. Skin physiological indicators (transepidermal water loss, erythema index, stratum corneum hydration, pH, temperature) of masked and unmasked areas were collected at baseline, after three hours of mask-wearing, post-tape stripping, and after 24 h, respectively. Microbiome samples collected from the masked areas before and after wearing a medical mask were analyzed with bioinformatics methods. Mask-wearing significantly weakened barrier function in both masked and adjacent unmasked areas, while reducing bacterial diversity. It was also associated with an increase in Cutibacterium (P = 0.110) and decreases in Streptococcus (P = 0.032) and Prevotella (P = 0.026) abundance. Moisturizer application prior to mask-wearing significantly reduced transepidermal water loss and erythema (both P < 0.001) and further improved erythema after 24 h (P = 0.048). These findings demonstrate that mask-wearing can disrupt the skin barrier and microbiome in individuals with sensitive skin and applying a moisturizer beforehand can mitigate mask-related discomforts by aiding barrier repair and reducing sensitivity.

Mitigating Adversarial Attacks in Object Detection through Conditional Diffusion Models

The field of object detection has witnessed significant advancements in recent years, thanks to the remarkable progress in artificial intelligence and deep learning. These breakthroughs have significantly enhanced the accuracy and efficiency of detecting and categorizing objects in digital images. Nonetheless, contemporary object detection technologies have certain limitations, such as their inability to counter white-box attacks, insufficient denoising, suboptimal reconstruction, and gradient confusion. To overcome these hurdles, this study proposes an innovative approach that uses conditional diffusion models to perturb adversarial examples. The process begins with the application of a random chessboard mask to the adversarial example, followed by the addition of a slight noise to fill the masked area during the forward process. The adversarial image is then restored to its original form through a reverse generative process that only considers the masked pixels, not the entire image. Next, we use the complement of the initial mask as the mask for the second stage to reconstruct the image once more. This two-stage masking process allows for the complete removal of global disturbances and aids in image reconstruction. In particular, we employ a conditional diffusion model based on a class-conditional U-Net architecture, with the source image further conditioned through concatenation. Our method outperforms the recently introduced HARP method by 5% and 6.5% in mAP on the COCO2017 and PASCAL VOC datasets, respectively, under non-APT PGD attacks. Comprehensive experimental results confirm that our method can effectively restore adversarial examples, demonstrating its practical utility.

Visual defect obfuscation based self-supervised anomaly detection

Due to scarcity of anomaly situations in the early manufacturing stage, an unsupervised anomaly detection (UAD) approach is widely adopted which only uses normal samples for training. This approach is based on the assumption that the trained UAD model will accurately reconstruct normal patterns but struggles with unseen anomalies. To enhance the UAD performance, reconstruction-by-inpainting based methods have recently been investigated, especially on the masking strategy of suspected defective regions. However, there are still issues to overcome: (1) time-consuming inference due to multiple masking, (2) output inconsistency by random masking, and (3) inaccurate reconstruction of normal patterns for large masked areas. Motivated by this, this study proposes a novel reconstruction-by-inpainting method, dubbed Excision And Recovery (EAR), that features single deterministic masking based on the ImageNet pre-trained DINO-ViT and visual obfuscation for hint-providing. Experimental results on the MVTec AD dataset show that deterministic masking by pre-trained attention effectively cuts out suspected defective regions and resolves the aforementioned issues 1 and 2. Also, hint-providing by mosaicing proves to enhance the performance than emptying those regions by binary masking, thereby overcomes issue 3. The proposed approach achieves a high performance without any change of the model structure. Promising results are shown through laboratory tests with public industrial datasets. To suggest EAR be possibly adopted in various industries as a practically deployable solution, future steps include evaluating its applicability in relevant manufacturing environments.

Preventing unwanted atomic layer deposition by liquid sealing

Atomic layer deposition (ALD) is an essential thin film fabrication technique widely used in electronic and energy systems, but avoiding ALD in untargeted areas has been a long-standing challenge that excludes the possibility of creating patterns through bottom-up processes. We present a liquid sealing strategy to prevent unwanted ALD by combining shadow masks with viscous perfluoropolyether oil that conformally fulfills all gaps between the mask and substrate down to the molecular level. Due to the anti-adsorption and non-defective nature of the liquid sealant, ALD molecules are forced away from the masking areas, resulting in order-of-magnitude improvements in the patterning resolution compared with conventional shadow masks. The liquid sealant does not change the growth rate and film properties of ALD oxides and can be easily removed by solvents. This liquid sealing strategy applies to a variety of shadow masks (e.g., curved mask and anodic aluminum oxide), substrates (e.g., silicon, plastic, and elastomer), patterning materials (e.g., oxide and metal), and deposition methods (e.g., sputtering and thermal evaporation), providing new insights for bottom-up pattern fabrications.

Extendable machine tool wear monitoring process using image segmentation based deep learning model and automatic detection of depth of cut line

Automating the monitoring of machine tools poses a significant challenge, with previous studies relying on machine vision to address this issue, primarily focusing on measuring specific tool wear using custom algorithms. In this study, we introduce an automated tool wear monitoring process capable of assessing wear across various tools. Utilizing deep neural network models for image classification and segmentation, our proposed process effectively masks areas where tool wear occurs. The image classification model selects the best-performing backbone model based on the highest F1 score. To account for varying depths-of-cut lines among different tools in the masked area, we introduce an algorithm that utilizes the Hough transform to determine the horizontal angle with the cut line. By systematically measuring the maximum flank wear of diverse tools, including indexable, ball, and solid endmills, using this method, we can replace conventional manual processes. This approach can be extended and automated for various machine tools, offering substantial potential for enhancing manufacturing tool monitoring processes.

Analysis of Bottom Reverberation Intensity Under Beam-Controlled Emission Conditions in Deep Water

A comprehensive understanding of the characteristics and the formation mechanism of reverberation is the key to improving the performance of the active target detection. In response to the challenge of analyzing the intensity of bottom reverberation in typical deep-sea environments, this study proposes a prediction method for the bottom reverberation intensity under beam-controlled emission conditions. It explains the variation law of bottom reverberation intensity under beam-controlled emission conditions in typical deep-sea environments of the South China Sea through theoretical and simulation analyses. Reverberation intensity of the deep-sea bottom under beam-controlled emission conditions exhibits significant fluctuations during the duration of reverberations in the direct sound zone of the seabed. This phenomenon is closely related to the directionality of the source emission, leading to intermittent reverberation masking and detectable areas in the active sonar detection. In addition, the duration of the high-reverberation zone near the cutoff distance of the direct sound from the seabed is longer under the beam-controlled emission conditions of the emission array located within the surface waveguide layer of the deep sea during winter.

Spatial and Temporal Variation of Subseasonal-to-Seasonal (S2S) Precipitation Reforecast Skill across the CONUS

Abstract Precipitation forecasts, particularly at subseasonal-to-seasonal (S2S) time scale, are essential for informed and proactive water resource management. Although S2S precipitation forecasts have been evaluated, no systematic decomposition of the skill, Nash–Sutcliffe efficiency (NSE) coefficient, has been analyzed toward understanding the forecast accuracy. We decompose the NSE of S2S precipitation forecast into its three components—correlation, conditional bias, and unconditional bias—by four seasons, three lead times (1–12, 1–22, and 1–32 days), and three models, European Centre of Medium-Range Weather Forecasts (ECMWF), National Centers for Environmental Prediction’s (NCEP) Climate Forecast System (CFS) model, and Environment and Climate Change Canada (ECCC), over the conterminous United States (CONUS). Application of a dry threshold, removal of grid cells with seasonal climatological precipitation means below 0.01 in. per day, is important as the NSE and correlations are lower across all seasons after masking areas with low precipitation values. Further, a west-to-east gradient in S2S forecast skill exists, and forecast skill was better during the winter months and for areas closer to the coast. Overall, ECMWF’s model performance was stronger than both ECCC and NCEP CFS’s performance, mainly for the forecasts issued during the fall and winter months. However, ECCC and NCEP CFS performed better for the forecast issued during the spring months and for areas further from the coast. Postprocessing using simple model output statistics could reduce both unconditional and conditional biases to zero, thereby offering better skill for regimes with high correlation. Our decomposition results show that efforts should focus on improving model parameterization and initialization schemes for climate regimes with low correlation.

Quality Control of Doppler Spectra from a Vertically Pointing, S-Band Profiling Radar

Abstract This study describes a novel combination of methods to remove spurious spectral peaks, or “spurs,” from Doppler spectra produced by a vertically pointing, S-band radar. The University of Massachusetts S-band frequency-modulated, continuous-wave radar (UMass FMCW) was deployed to monitor the growth of the CBL over northern Alabama during the VORTEX–Southeast field campaign in 2016. The Doppler spectra contained spurs caused by high-voltage switching power supplies in the traveling wave tube amplifier. In the original data-processing scheme for this radar, a median filtering method was used to eliminate most of the spurs, but the largest ones persisted, which significantly degraded the quality of derived radar moments (e.g., reflectivity, Doppler velocity, and spectrum width) and hindered further analysis of these data (e.g., hydrometeor classification and boundary layer height tracking). Our technique for removing the spurs consists of three steps: (i) a Laplacian filter identifies and masks peaks in the spectra that are characteristic of the spurs in shape and amplitude, (ii) an in-painting method then fills in the masked area based on surrounding data, and (iii) the moments data (e.g., reflectivity, Doppler velocity, and spectrum width) are then recomputed using a coherent power technique. This combination of techniques was more effective than the median filter at removing the largest spurs from the Doppler spectra and preserved more of the underlying Doppler spectral structure of the scatterers. Performance of both the median-filter and the in-painting methods is assessed through statistical analysis of the spectral power differences. Downstream products, such as boundary layer height detection, are more easily derived from the recomputed moments. Significance Statement This manuscript describes a novel combination of image and signal processing techniques used to recover meteorological observations from corrupted Doppler radar spectra. This successful recovery of meteorologically significant information illustrates the importance of retaining Doppler spectra when practical. In seeking solutions to data quality issues, the atmospheric science community should remain cognizant of promising techniques offered by other disciplines. We present this data rescue study as an example to the meteorological community.

Jazan's Catch Chronicles: A Statistical Exploration of Lethrinus Nebulous and Acanthocybium Solandri Fishing (2012-2022)

This study delves into the fascinating world of fisheries data, specifically focusing on two commercially important fish species in the Red Sea: The Lethrinus nebulous (cloud emperor) and the Acanthocybium solandri (wahoo). Through the lens of statistics, this study aims to uncover valuable insights into the population dynamics, catch patterns and potential management strategies for these fish species. By analyzing data from "Jazan's Catch Chronicles," from General Administration of Fisheries Resources (Ministry of Environment, Water and Agriculture) -the Statistical Yearbook 2022-, we employ statistical methods to extract meaningful patterns from the data and answer critical questions. This knowledge can then inform fisheries management practices, ensuring the long-term sustainability of these valuable resources. The findings suggest that the diversity of the catch may have increased slightly over the study period. The number of species with a catch of more than 100 tonnes increased from 5 in 2017 to 7 in 2022. The number of species with a catch of more than 10 tonnes also increased from 14 in 2017 to 17 in 2022. However, further statistical analysis, such as a chi-squared test, is needed to determine if this difference is statistically significant. The study also highlights the importance of balancing data disclosure with species conservation. While detailed location data can be valuable for scientific research, it can also pose threats to endangered species. In the case of the IUCN-listed Cheilinus undulatus (humphead wrasse), a decision tree approach was employed to weigh the risks and rewards of disclosing location data. Considering the species' endangered status and declining population, the study recommends restricting data by masking species IDs and areas or restricting information by publishing high-resolution habitat maps without specific location details. This approach helps balance the need for scientific knowledge with protecting vulnerable species. Overall, this study sheds light on the dynamics of fish populations in the Red Sea, emphasizing the need for sustainable management practices and responsible data handling to ensure the long-term health of these vital marine resources.

Automatic Segmentation of Membranous Glottal Gap Area with U-Net-Based Architecture.

While videostroboscopy is recognized as the most popular approach for investigating vocal fold function, evaluating the numerical values, such as the membranous glottal gap area, remains too time consuming for clinical applications. We used a total of 2507 videostroboscopy images from 137 patients and developed five U-Net-based deep-learning image segmentation models for automatic masking of the membranous glottal gap area. To further validate the models, we used another 410 images from 41 different patients. During development, all five models exhibited acceptable and similar metrics. While the VGG19 U-Net had a long inference time of 1654 ms, the other four models had more practical inference times, ranging from 16 to 138 ms. During further validation, Efficient U-Net demonstrated the highest intersection over union of 0.8455, the highest Dice coefficient of 0.9163, and the lowest Hausdorff distance of 1.5626. The normalized membranous glottal gap area index was also calculated and validated. Efficient U-Net and VGG19 U-Net exhibited the lowest mean squared errors (3.5476 and 3.3842) and the lowest mean absolute errors (1.8835 and 1.8396). Automatic segmentation of the membranous glottal gap area can be achieved through U-net-based architecture. Considering the segmentation quality and speed, Efficient U-Net is a reasonable choice for this task, while the other four models remain valuable competitors. The models' masked area enables possible calculation of the normalized membranous glottal gap area and analysis of the glottal area waveform, revealing promising clinical applications for this model. NA Laryngoscope, 134:2835-2843, 2024.

An Approach for Brick Wall Quantity Take-Off by U-Net Method Based on Deep Learning

This study presents a deep learning-based method for the quantity take-off in the construction industry. In this context, the brick wall quantity calculation was performed automatically over two-dimensional (2D) pictures by the U-Net method. 280 photos were first taken in the field at different distances and angles. 1960 images were, then, obtained by augmentation to increase the training accuracy. Pixel calculation of the automatically masked area in the images was made for wall estimation. The wall area was calculated by comparing this pixel value with that of the reference brick surface area. The method was tested on four sample photos including different wall images. A parametric study was carried out to reduce the errors. In the study, it has been shown that the proposed method is suitable for brick quantity calculation. In addition, it was concluded that the photo should be taken as close as possible, and more than one brick should be taken as a reference in close-up photos to increase the accuracy.

Atomic Layer Etching of GaN and AlGaN Using CH4/H2, H2 and HCl Chemistry

Modern microelectronic components for high-power and high-frequency applications require reliability and reproducibility of the production processes. Atomic layer etching (ALE) represents a key technology for achieving these requirements.An ALE sequence is characterized by a chemical modification step that affects only the top atomic layer of the surface (adhesion) and an ion assisted non-reactive inert gas etching step that removes only this chemically modified area due to creation of volatile byproducts. These two steps are separated by inert gas purge. This grants the etching of single atomic layers. The etching step can be triggered by thermal energy or kinetic energy by impinging particles (e.g. ion bombardment by a plasma). The low energy required for ALE guarantees low-damage etching.In literature ALE processes using Cl2 are often reported. In this work, the results of plasma-enhanced ALE of GaN and AlGaN using different chemistries of CH4/H2, H2 and HCl as etch gases are presented. The experiments were performed with a highly customized etch facility equipped with a capacitively coupled plasma source (CCP, 60 MHz) and a substrate bias generator (RF, 2 MHz). The investigations were carried out on epitaxially deposited GaN on sapphire with masks of protective resist and SiO2 as well as GaN-AlGaN heterostructures with patterned SiO2 maskings. For process analysis optical emission spectroscopy (OES) was carried out. The etching rate and roughness were determined by ellipsometry and atomic force microscopy (AFM) as well as scanning electron microscopy (SEM).Observations by OES indicate a successful excitation and dissociation of the etching molecules. With rising plasma powers physical sputtering effects were revealed. At the beginning of the process development continuous etching processes were examined. With CH4/H2 chemistry using GaN samples it was found that the etching behavior is strongly influenced by the type of masking material. Protective resist resulted in a low etch rate of 7.5 nm/min and significant damage of the masking material. The cause of this behavior is assumed to be redeposition of material by interaction of methane, resist, and plasma. With the change to a hard mask of SiO2 etch rates of 45 to 72 nm/min were achieved, increasing with RF-power and CH4/H2 ratio. Adding methane led to detrimental side effects such as dendrite growth and wall formation on the transient region of unmasked to former masked sample areas. In the further development of the ALE process an etch chemistry using only hydrogen as the chemical etchant was used resulting in an etch per cycle (EPC) of about 3 Å for GaN samples. Further investigations of ALE with hydrogen etch chemistry on GaN-AlGaN samples revealed that AlGaN, in contrast to pure GaN, could not be etched successfully with this chemistry.Substituting CH4/H2 or H2 to HCl enables the etch of AlGaN. The investigations carried out indicate that this etching process is also a self-limiting ALE. AFM measurements revealed an EPC of 1.2 Å for lower RF-powers (50 W). Increasing the RF power to 150 W only slightly enhance the EPC to 2.0 Å. The surface roughness tends to decrease with rising RF-power from 50 to 300 W.

MTSA-Net: A multiscale time self-attention network for ship radiated self-noise reduction

For a surface ship carrying a towed array, its in-band radiated self-noise is one of the near-field strong interference, which will seriously limit the performance of the underwater acoustic (UWA) signal detection system. Typically, the conventional ship noise cancellation methods requires time-synchronized hydrophone arrays, such as the towed linear array (TLA), to suppress its self-noise by using spatial filters. However, the spatial filter based methods fail when the direction of arrival of the desired signal is in the ship noise masking area. In cooperative scenarios, the prior knowledge of the template signal provides additional temporal information, which can be utilized to design a time-domain representations based detection system. In this paper, a multiscale time self-attention network (MTSA-Net) is proposed to mitigate the ship radiated self-noise and enhance the desired signal to improve the performance of signal detection system. Experimental results based on sea trial data indicate the effectiveness of our proposed method.

Automatic segmentation of atrial fibrillation and flutter in single-lead electrocardiograms by self-supervised learning and Transformer architecture.

Automatic detection of atrial fibrillation and flutter (AF/AFL) is a significant concern in preventing stroke and mitigating hemodynamic instability. Herein, we developed a Transformer-based deep learning model for AF/AFL segmentation in single-lead electrocardiograms (ECGs) by self-supervised learning with masked signal modeling (MSM). We retrieved data from 11 open-source databases on PhysioNet; 7 of these databases included labeled ECGs, while the other 4 were without labels. Each database contained ECG recordings with durations of ≥30s. A total of 24 intradialytic ECGs with paroxysmal AF/AFL during 4h of hemodialysis sessions at Seoul National University Hospital were used for external validation. The model was pretrained by predicting masked areas of ECG signals and fine-tuned by predicting AF/AFL areas. Cross-database validation was used for evaluation, and the intersection over union (IOU) was used as a main performance metric in external database validation. In the 7 labeled databases, the areas marked as AF/AFL constituted 41.1% of the total ECG signals, ranging from 0.19% to 51.31%. In the evaluation per ECG segment, the model achieved IOU values of 0.9254 and 0.9477 for AF/AFL segmentation and other segmentation tasks, respectively. When applied to intradialytic ECGs with paroxysmal AF/AFL, the IOUs for the segmentation of AF/AFL and non-AF/AFL were 0.9896 and 0.9650, respectively. Model performance by different training procedure indicated that pretraining with MSM and the application of an appropriate masking ratio both contributed to the model performance. It also showed higher IOUs of AF/AFL labels than in previous studies when training and test databases were matched. The present model with self-supervised learning by MSM performs robustly in segmenting AF/AFL.

Anomaly-prior guided inpainting for industrial visual anomaly detection

Visual anomaly detection aims to identify areas where the appearance deviates from the normal distribution. Reconstruction-based methods detect anomalies through the analysis of reconstruction errors. In particular, Inpainting-based methods aim to reduce the potential for network generalization of anomalies by introducing random masks. However, larger anomalies may still be reconstructed, making them challenging to discriminate. Although enhancing the network’s long-distance modeling capability offers the possibility to mitigate the reconstruction of anomalous regions, there is a possibility that the reconstruction error for normal regions increases and the ability to discriminate small anomalies is reduced due to the random nature of the mask placement. To address this issue, we introduce a correction branch to modify the original reconstruction results obtained from the reconstruction network during testing. This method can alleviate the potential increase in reconstruction errors in normal regions caused by the randomness of the mask to some extent and enhance the network’s responsiveness to anomalous regions. We also propose a reconstruction network constructed with ConvNeXt blocks and the incorporation of the channel attention mechanism, which enhances the long-distance modeling ability, thereby improving the model’s predictive capacity for large masked areas, this network also improves the reconstruction of textures, thereby improving the network’s capacity to discriminate anomalies. The proposed method achieves competitive results on the challenging benchmark MVTec AD and BTAD datasets.

Geo-indistinguishable masking: enhancing privacy protection in spatial point mapping

ABSTRACT Spatial point mapping is a useful practice in exploratory point pattern analysis, but it poses significant privacy risks as the identity of individuals may be revealed from the maps. Geomasking methods have been developed to mitigate the risks by displacing spatial points before mapping. However, many of these methods rely on a weak privacy notion called spatial k-anonymity, which is insufficient to withstand the growing amount of spatial data (e.g. land use) that adversaries can use as side information to infer the actual locations of individuals. We proposes a method called geo-indistinguishable masking to address this issue by relying on a strong privacy notion called geo-indistinguishability. This notion ensures consistent levels of privacy protection regardless of any side information. The method consists of two steps. The first step involves creating a masking area for each spatial point to include a set of candidate locations to which the point can be relocated. In the second step, we formulate an optimization model to ensure the masked locations satisfy geo-indistinguishability while minimizing the distance displaced. Computational experiments on a synthetic dataset demonstrate that our proposed method is both efficient and effective in providing strong privacy protection while preserving the spatial point patterns.

RemainNet: Explore Road Extraction from Remote Sensing Image Using Mask Image Modeling

Road extraction from a remote sensing image is a research hotspot due to its broad range of applications. Despite recent advancements, achieving precise road extraction remains challenging. Since a road is thin and long, roadside objects and shadows cause occlusions, thus influencing the distinguishment of the road. Masked image modeling reconstructs masked areas from unmasked areas, which is similar to the process of inferring occluded roads from nonoccluded areas. Therefore, we believe that mask image modeling is beneficial for indicating occluded areas from other areas, thus alleviating the occlusion issue in remote sensing image road extraction. In this paper, we propose a remote sensing image road extraction network named RemainNet, which is based on mask image modeling. RemainNet consists of a backbone, image prediction module, and semantic prediction module. An image prediction module reconstructs a masked area RGB value from unmasked areas. Apart from reconstructing original remote sensing images, a semantic prediction module of RemainNet also extracts roads from masked images. Extensive experiments are carried out on the Massachusetts Roads dataset and DeepGlobe Road Extraction dataset; the proposed RemainNet improves 0.82–1.70% IoU compared with other state-of-the-art road extraction methods.