Balancing accuracy and efficiency in particle tracking: analyzing image resolution and batch size trade-offs

There is a need for video encoders to generate bit streams of quality that matches human evaluation. This is often achieved by adopting perceptual quality metrics in video encoders. A huge body of work has been devoted to the development of perceptual video quality metrics that take the properties of human visual system into account. It is our interest in this paper to study how the existing video coding systems can benefit from such perceptual quality metrics. Specifically, we examine the existing perceptual video coding systems and introduce the concept of i§codec-friendliness, i¨ meaning how the perceptual quality metrics can be nicely incorporated into the coding process of standard video coding algorithms. We conclude the study by suggesting guidelines for the development of future quality metrics from the video coding perspective.

Delivering high perceptual quality video over wireless channels is challenging due to the changing channel quality and the variations in the importance of one source packet to the next for the end-user's perceptual experience. Leveraging perceptual metrics in concert with link adaptation to maximize perceptual quality and satisfy real-time delay constraints is largely unexplored. We introduce an APP/MAC/PHY cross-layer architecture that enables optimizing perceptual quality for delay-constrained scalable video transmission. We propose an online QoS-to-QoE mapping technique to quantify the loss visibility of packets from each video layer using the ACK history and perceptual metrics. At the PHY layer, we develop a link adaptation technique that uses the QoS-to-QoE mapping to provide perceptually-optimized unequal error protection per layer according to packet loss visibility. At the APP layer, the source rate is adapted by selecting the set of temporal and quality layers to be transmitted based on the channel statistics, source rates, and playback buffer state. The proposed cross-layer optimization framework allows the channel to adapt at a faster time scale than the video codec. Furthermore, it provides a tradeoff between playback buffer occupancy and perceptual quality. We show that the proposed architecture prevents playback buffer starvation, provides immunity against short-term channel fluctuations, regulates the buffer size, and achieves a 30% increase in video capacity versus throughput-optimal link adaptation.

Perceptual image quality metrics have explicitly accounted for perceptual characteristics of the human visual system (HVS) by modeling sensitivity to subband noise as the just-noticeable threshold of distortion. While such metrics can successfully account for contrast and luminance masking, they are quite sensitive to spatial shifts, intensity shifts, contrast changes, and scale changes. In contrast, the recently proposed structural similarity (SSIM) metrics account for perception more implicitly with the assumption that the HVS is adapted for extracting structural information (relative spatial covariance) from images. As such, they have the potential to be much more effective in quantifying suprathreshold compression artifacts than traditional perceptual metrics, as such artifacts tend to distort the structure of an image. We use a (perceptually) weighted variation of the complex wavelet SSIM (CWSSIM) to evaluate standard image compression techniques such as JPEG, JPEG 2000, SPIHT, and the Safranek-Johnston perceptual image coder. Our experimental results indicate that the weighted CWSSIM generally agrees with subjective evaluations.

<p class="MsoNormal" style="mso-pagination:none;mso-layout-grid-align:none; text-autospace:none">This thesis investigates the behaviour of the often used point-wise skill score, the MSESSini a.k.a. BSS, and develops new error metrics that, as opposed to point-wise metrics, take the spatial structure of morphological patterns into account. The MSESSini measures the relative accuracy of a morphological prediction over a prediction of zero morphological change, using the mean-squared error (MSE) as the accuracy measure. The main findings about the MSESSini are: 1) a generic ranking, based on values for MSESSini, has limited validity, since the zero change reference model fails to make model performance comparable across different prediction situations; 2) the combination of larger, persistent and smaller, intermittent scales of cumulative change may lead to an increase of skill with time, without the prediction on either of these scales becoming more skilful with time; 3) in the presence of inevitable location errors, the MSESSini favours predictions that underestimate the variance of cumulative bed changes and 4) existing methods to correct for measurement error are inconsistent in either their skill formulation or their suggested classification scheme. In order to overcome the inherent limitations of point-wise metrics, three novel diagnostic tools for the spatial validation of 2D morphological predictions are developed. First, a field deformation or warping method deforms the predictions towards the observations, minimizing the squared point-wise error. Error measures are formulated based on both the smooth displacement field between predictions and observations and the residual point-wise error field after the deformation. In contrast with the RMSE, the method captures the visual closeness of morphological patterns. Second, an optimal transport method defines the distance between predicted and observed morphological fields in terms of an optimal sediment transport field. The optimal corrective transport field moves the misplaced sediment from the predicted to the observed morphology at the lowest quadratic transportation cost. The root-mean-squared value of the optimal transport field, the RMSTE, is proposed as a new error metric. As opposed to the field deformation method, the optimal transport method is mass-conserving, parameter-free and symmetric. The RMSTE, unlike the RMSE, is able to discriminate between predictions that differ in the misplacement distance of predicted morphological features. It also avoids the consistent reward of the underestimation of morphological variability that the RMSE is prone to. Third, a scale-selective validation approach allows any metric to selectively address multiple spatial scales. It employs a smoothing filter in such a way that, in addition to the domain-averaged statistics, localized validation statistics and maps of prediction quality are obtained per scale (geographic extent or areal size of focus). The employed skill score weights how well the morphological structure and variability are simulated, while avoiding to reward the underestimation of variability. To fully describe prediction quality multiple metrics are required with a weighting determined by the goal of the simulation. Point-wise metrics should be supplemented with an error decomposition, as to avoid undesired underestimation of variability. Further, a set of performance metrics must include a metric, e.g. the RMSTE, that accounts for the spatial structure of the observed and predicted morphological fields.</p>

Simulating solute transport in heterogeneous porous media poses computational challenges due to the high-resolution meshing required for traditional solvers. To overcome these challenges, this study explores a mesh-free method based on deep learning to accelerate solute transport simulation. We employ Physics-informed Neural Networks (PiNN) with a periodic activation function to solve solute transport problems in both homogeneous and heterogeneous porous media governed by the advection-dispersion equation. Unlike traditional neural networks that rely on large training datasets, PiNNs use strong-form mathematical models to constrain the network in the training phase and simultaneously solve for multiple dependent or independent field variables, such as pressure and solute concentration fields. To demonstrate the effectiveness of using PiNNs with a periodic activation function to resolve solute transport in porous media, we construct PiNNs using two activation functions, sin and tanh, for seven case studies, including 1D and 2D scenarios. The accuracy of the PiNNs’ predictions is then evaluated using absolute point error and mean square error metrics and compared to the ground truth solutions obtained analytically or numerically. Our results demonstrate that the PiNN with sin activation function, compared to tanh activation function, is up to two orders of magnitude more accurate and up to two times faster to train, especially in heterogeneous porous media. Moreover, PiNN’s simultaneous predictions of pressure and concentration fields can reduce computational expenses in terms of inference time by three orders of magnitude compared to FEM simulations for two-dimensional cases.

Blocking artifact reduction or deblocking algorithm is an important component in modern block-based video encoding architecture and often used as post-processing procedures in many encoding/transcoding applications. Most of the existing video deblocking algorithms do not take into account Human Visual System(HVS) models and employ empirically designed filters, resulting in suboptimal perceptual image quality and the difficulty to adjust parameters for different application scenarios. This paper presents a new deblocking algorithm that automatically generates spatially variant adaptive filters by maximizing a regularized perceptual quality metric. Such optimization based approach leads to better human perceived image quality and much easier filter parameter adjustment.

Total electron content (TEC) serves as a key parameter characterizing ionospheric conditions. Accurate prediction of TEC plays a crucial role in improving the precision of Global Navigation Satellite Systems (GNSS). However, existing research have predominantly emphasized spatial variations in the ionosphere, neglecting the periodic changes of the ionosphere with the diurnal cycle. In this paper, we propose a TEC prediction model, which simultaneously considers both spatial and temporal characteristics to extract spatiotemporal features of ionospheric distribution. Additionally, we integrate several space weather element datasets into the prediction model framework, allowing the generation of multiple space weather feature values that represent the influence of space weather on the ionosphere at different latitudes and longitudes. Moreover, we apply Gaussian process regression (GPR) interpolation to geomagnetic data to characterize impact on the ionosphere, thereby enhancing the prediction accuracy. We compared our model with traditional image-based models such as convolutional neural networks (CNNs), convolutional long short-term memory networks (ConvLSTMs), a self-attention mechanism-integrated ConvLSTM (SAM-ConvLSTM) model, and one-day predicted ionospheric products (C1PG) provided by the Center for Orbit Determination in Europe (CODE). We also examined the effect of using different numbers of space weather feature values in these models. Our model outperforms the comparison models in terms of prediction error metrics, including mean absolute error (MAE), root mean square error (RMSE), correlation coefficient (CC), and the structural similarity index (SSIM). Furthermore, we analyzed the influence of different batch sizes on model training accuracy to find the best performance of each model. In addition, we investigated the model performance during geomagnetic quiet periods, where our model provided the most accurate predictions and demonstrates higher prediction accuracy in the equatorial anomaly region. We also analyzed the prediction performance of all models during space weather events. The results indicate that the proposed model is the least affected during geomagnetic storms and demonstrates superior prediction performance compared to other models. This study presents a more stable and high-performance spatiotemporal prediction model for TEC.

Stocks have become one of the largest and most intricate financial markets globally due to their high popularity, making them very challenging to predict as they can process millions of transactions rapidly. The objective of this study is to enhance the field by creating a dependable and accurate model for predicting the stock price of Amazon. This will be achieved via the use of advanced algorithms such as Long Short-Term Memory (LSTM) and Gated Recurrent Unit (GRU). This research utilised historical data on Amazon's stock price from the past five years, which was acquired from Yahoo Finance. The data was partitioned using a hold-out validation technique, allocating 80% for training and 20% for testing. The model underwent training using different optimizers (Adam, SGD, RMSprop), batch sizes (8, 16, 32), and learning rates (0.001, 0.0001). The evaluation criteria comprised of mean squared error (MSE), root mean squared error (RMSE), mean absolute error (MAE), and mean absolute percentage error (MAPE). The results suggest that the GRU model, when trained with the RMSprop optimizer using a batch size of 16 and a learning rate of 0.0001, as well as with the SGD optimizer using a batch size of either 16 or 32 and a learning rate of either 0.001 or 0.0001, produced the lowest error metrics, indicating superior performance. This study enables more precise forecasts of stock prices and more efficient investment techniques.

This paper proposes the development and evaluation of an auto-tuning stochastic resonance (SR) for image enhancement on images under various illumination conditions. Perceptual Quality Metric (PQM) is used for evaluating and quantifying the image quality. The current process is developed being based on our previous works related to the image enhancement by using the manual tuning stochastic resonance. The process was performed by adding the random noise and threshold in an image. The process works properly in the dark and very low contrast images as well as bright images. This image enhancement system works on dark and bright images as well. The system was tested with the face detection algorithm on the dark and illumination variant images. In this paper, we present the idea of auto-tuning of the SR iteration with random noise and threshold value 0 by using the process related to the histogram calculation, mean and median. In this paper, we performed various experiments on object and human detection as well under different conditions and confirmed the effectiveness of our auto-tuning SR based image enhancement algorithm. Finally, we conducted experiments using Perceptual Quality Metric which is an image quality metric to apply the proposed algorithm in various fields.

Accurate modeling of wind speed distributions is a crucial step in the design and sizing of small wind turbines. However, most current studies rely solely on the two-parameter Weibull distribution, without rigorous statistical analysis or comparison with other methods. This study presents a statistical analysis for the sizing of a small horizontal-axis wind turbine in northern Morocco. The proposed framework integrates hub-height optimization, probabilistic wind distribution modeling, and turbine parameter computation within a unified and robust methodology. The site hourly wind speed data, obtained from PVGIS, were extrapolated to an optimal hub height of 29 m using a power-law formulation combined with a cost-energy optimization approach. Then multiple distribution models were evaluated, which are the Weibull, Champernowne distribution, and kernel density estimation (KDE). Their performance was assessed using both conventional pointwise error metrics (RMSE and MAE) and distributional divergence measures, namely, Jensen‒Shannon divergence (JSD) and the Wasserstein distance (WD). Although the least-squares Weibull model produced the lowest pointwise errors (RMSE = 0.0136, MAE = 0.0103), it exhibited some distributional misspecification (JSD = 0.0132, WD = 0.0307) concentrated in the upper tail. Finally, based on a piecewise power-curve integration, the resulting turbine design parameters correspond to a swept area of 37.82 m², a blade length of 3.47 m, and a rated power of 8.28 kW to meet an annual energy demand of 33159 kWh. The results demonstrate that distributional metrics reveal modeling biases in the Weibull approach that remain undetected by traditional pointwise evaluation methods, with important implications for accurate energy yield prediction and turbine sizing reliability.

Groundwater appearance on the land surface in the Sphinx and Pyramids area in Egypt prompted a study to identify the sources of this rising groundwater and propose short term solutions to the problem. Although a short-term solution was designed and implemented in the site, concerns remained regarding the potential interactions between the leaking water (mostly sewage water) and the limestone aquifer underlying Sphinx and the Pyramids area. These interactions may lead to widening the fracture and fault openings, thereby increasing their permeability, increasing the volumes of groundwater passing through them and affecting the stability of the archaeological structures. In this study, a two dimensional numerical model is developed to investigate the effect of limestone dissolution on groundwater fluxes entering the subsurface domain underneath the Sphinx and Pyramids area. The model generates zones of high hydraulic conductivities to represent fractures and the hydraulic conductivity field coupled with appropriate boundary conditions is used to solve the flow equation to obtain the potential field and subsequently the velocity field. The particle tracking approach is adapted to solve the transport and dissolution reactions problems. The model considered the rate of dissolution reactions at mineral surface, mass transfer of reaction products between the mineral surface and bulk fluid, and reactions involving the distribution of CO2 gas in water as the three potential limiting conditions for dissolution. Comparisons between the case with dissolution and that without the effect of dissolution reveal the importance of the process.

The success of Internet of Things (IoT) has significantly increased the volume of data generated by various smart applications. However, as many of these applications are characterized by strict Quality of Service (QoS) requirements, there is a growing need for accurately predicting typical performance parameters such as throughput. This prediction should be based on the applications' traffic profiles and at the same time reflect the network uncertainty that IoT access networks add to the overall communication. In this work, we deployed 6 different smart building applications in a real testbed while creating a considerable traffic contention in an IEEE 802.15.4 access network. After preprocessing the raw data and following a feature engineering mechanism, we apply five different regression learning approaches to each application and predict its throughput. By resorting to several prediction error metrics and time metrics such as training and inference time, we show that the multiple linear regression achieves high accuracy while outperforming other well known machine learning methods.

Recent advances in generative models have enabled the synthesis of high-fidelity images across various domains. However, applications in mental health remain underexplored, particularly in the context of interactive emotional visualization. This paper introduces GOOD (GAN Optimized with One-step Diffusion), a hybrid generative framework that integrates a GAN-based refinement mechanism into a one-step diffusion model. The proposed approach addresses key limitations of current one-step methods by enabling localized correction of visual artifacts while preserving low-latency generation. Experimental results demonstrate improvements of up to +32 dB in PSNR (indicating sharper and less noisy outputs) and +0.9 in SSIM (reflecting better structural fidelity and perceptual quality) over the one-step baseline. Beyond quantitative gains, the enhanced visual clarity and stability are particularly valuable in mindfulness and therapeutic settings, where more coherent and emotionally resonant imagery can foster user engagement, reduce distraction caused by artifacts, and support mental well-being through more immersive visualization experiences.

In traditional time-series image prediction tasks, both accuracy and image quality tend to deteriorate as the prediction horizon extends. To address this challenge in Shiitake mushroom fruiting body growth prediction, this study selected Shiitake mushroom strain No. 509, cultivated by the Shanghai Academy of Agricultural Sciences, as the experimental subject and proposed an enhanced model, VMRNN-DMSA, based on the Vision Mamba RNN Depth architecture. This model integrates a skip-connection mechanism with a Max Feature Map module to effectively filter and fuse features, enhancing feature representation and prediction accuracy. Additionally, a Spatial Attention Mechanism was introduced to strengthen the perception of key regions and improve spatial modeling. Furthermore, an Adaptive Kernel Convolution module with irregular context convolution kernels was incorporated to extract fine-grained local features and enhance visual quality. A weighted loss function was used to balance pixel-level accuracy, structural fidelity, and perceptual quality. This function combines Mean Squared Error Loss, Multi-Scale Structural Similarity, and Perceptual Loss. Experimental results showed that the proposed method achieved an MSE of 39.4255, an SSIM of 0.8579, and a PSNR of 22.0774. Compared with baseline models, MSE decreased by 29.05%, while SSIM and PSNR increased by 19.34% and 14.52%, respectively. These results indicate that VMRNN-DMSA significantly improves both prediction accuracy and image quality in long-term forecasting tasks, providing a reliable reference for the growth prediction of other edible fungi.

Low-light Image Enhancement (LLIE) aims to restore images captured under insufficient illumination to a normally lit appearance with adequate brightness, clear structures, and natural colors. However, existing methods still perform poorly in extremely dark conditions, where structural distortion, loss of details, amplified noise, and color imbalance frequently occur, making it difficult to maintain both structural fidelity and perceptual quality. To address these challenges, this paper proposes CA-YUVNet, a cross-attention-based low-light enhancement model. The model converts input images from RGB to YUV space to decouple luminance and chrominance, thereby suppressing noise propagation and color distortion commonly observed in low-light scenes. Furthermore, a cross-branch crossattention mechanism is introduced in the U and V branches to mitigate color shifts and chromatic artifacts. To further improve detail restoration and color consistency, a Bipolar Channel Attention Fusion (BCAF) module is designed, which integrates multi-scale convolution and deep feature fusion to enhance noise suppression and structural refinement. Experimental results on the LOL-v1, LOL-v2-real, and LOL-v2-synthetic datasets demonstrate that CA-YUVNet outperforms existing methods, validating its effectiveness and robustness.