Neighboring Pixels Research Articles

Improved ELBP descriptors for face recognition

In this work, the three novel descriptors are introduced for face recognition (FR) so-called sobel horizontal elliptical local binary pattern (SHELBP), sobel vertical elliptical local binary pattern (SVELBP) and sobel elliptical local binary pattern (SELBP). All three proposed descriptors are the extensions of the work proposed by Nguyen and Caplier (2012). Nguyen and Caplier (2012) proposed three descriptors for FR called as HELBP, VELBP and ELBP. In HELBP and VELBP, the horizontal neighbourhood pixels (aligned elliptically) and vertical neighbourhood pixels (aligned elliptically) are compared with centre pixel to produce their feature sizes and ELBP is the combined histogram extracted from both the descriptors. The performance of these descriptors are not effective under illumination variations (without pre-processing), as it is experimentally proved in this work. To compensate for that sobel operator is applied as image pre-processing before feature extraction is performed. The features extracted from sobel magnitude and directional gradients eliminates this problem very effectively.

An Adaptive PolSAR Trilateral Filter Based on the Mechanism of Scattering Consistency

Polarimetric Synthetic Aperture Radar (PolSAR) can record the backscattering information of ground objects by capturing the phase difference of echo signal in different polarization combinations. Due to the characteristics of coherent imaging, there is a lot of coherent speckle noise in PolSAR images. However, most of the current PolSAR image despeckling algorithms are difficult to balance the ability in aspects of speckle suppression, polarization information retention and edge structure preservation. This paper proposes an adaptive PolSAR trilateral filter (PTF) based on the mechanism of scattering consistency, which consists of following three steps. (1) The marker map of ground object scattering category is firstly obtained based on the dominant scattering mechanism determined by Freeman decomposition of the polarization covariance matrix. (2) Next, the neighborhood pixels with the same scattering mechanism as the center pixel are screened out within an adaptive filtering window. (3) A PolSAR trilateral filter is established through integration of the spatial proximity, polarization similarity and local structural characteristics of the center pixel and the screened pixels to effectively suppress the speckle noises. Experimental results upon the data of Radarsat-2 and AIRSAR demonstrate that PTF can not only suppress coherent speckle noise effectively without sacrificing the edge structure information, but also retain the unique polarization information of PolSAR image. In particular, compared with the traditional six widely used PolSAR despeckling algorithms, PTF have been significantly improved in terms of Equivalent Number of Looks (ENL) and Degree of difference (Dod) indexes by 2.49 times and 6 orders of magnitude, respectively.

Spectral–Spatial Reweighted Robust Nonlinear Unmixing for Hyperspectral Images Based on an Extended Multilinear Mixing Model

Recently, nonlinear unmixing algorithms have attracted special attention in hyperspectral image (HSI) processing. However, the inherent wavelength-dependent nonlinear intensity and noise effects in real HSIs are often overlooked, and the spatial information of HSIs has not been fully utilized in current studies. In this paper, we propose a spectral-spatial reweighted robust nonlinear unmixing algorithm to solve the above problems. First, a robust unmixing method is built on an extended multilinear mixing model (EMLM), which employs the vectorized nonlinear parameters to describe the nonlinear intensity varying along with spectral bands, and adopts the <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">l</i> <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2,1</sub> norm-based loss function to suppress the influence of noise. Second, to fully exploit the spectral-spatial information of HSIs, the nonlinear unmixing problem is reformulated with two regularizers. Specifically, a reweighted collaborative sparse regularizer is used to make the pixels in a superpixel-based neighborhood share the same subset of endmembers and have similar abundances because the neighboring pixels are usually composed of several materials in similar proportions, and a reweighted spectral total variation regularizer is utilized to improve the spectral-spatial smoothness of the vectorized nonlinear parameters by considering the local-region similarities of the nonlinear mixing effects. Finally, the constrained optimization problem is solved by the alternating direction method of multipliers (ADMM). Experimental results on simulated, semi-simulated, and real hyperspectral datasets demonstrate that the proposed method outperforms several state-of-the-art nonlinear unmixing methods.

Change Detection in SAR Images via Ratio-Based Gaussian Kernel and Nonlocal Theory

Compared with the synthetic aperture radar (SAR) image processing theory based on local neighborhood, the nonlocal theory is not limited to a local neighborhood of an image and has great potential in change detection of SAR images. In this study, an approach using ratio-based nonlocal information (RNLI) is proposed for change detection in multitemporal SAR images. First, the RNLI is extracted from a spatial–temporal nonlocal neighborhood where the similarity of two pixels in the nonlocal neighborhood is well characterized by the proposed ratio-based Gaussian kernel function. The parameters of RNLI: noise level and matching window size are adaptively determined to avoid the uncertainty of the change detection result caused by user experience. Second, the difference image is generated by using the RNLI and the ratio operator. Finally, the change map is obtained by segmenting the difference image with a threshold. Experiments conducted on two real datasets and two simulated datasets showed that the proposed method performed better than the other advanced change detection methods, which can better retain the edge information of the changed area while reducing the overall error of the change detection results.

Self-Supervised Monocular Depth Estimation With Multiscale Perception

Extracting 3D information from a single optical image is very attractive. Recently emerging self-supervised methods can learn depth representations without using ground truth depth maps as training data by transforming the depth prediction task into an image synthesis task. However, existing methods rely on a differentiable bilinear sampler for image synthesis, which results in each pixel in a synthetic image being derived from only four pixels in the source image and causes each pixel in the depth map to perceive only a few pixels in the source image. In addition, when calculating the photometric error between a synthetic image and its corresponding target image, existing methods only consider the photometric error within a small neighborhood of each single pixel and therefore ignore correlations between larger areas, which causes the model to tend to fall into the local optima for small patches. In order to extend the perceptual area of the depth map over the source image, we propose a novel multi-scale method that downsamples the predicted depth map and performs image synthesis at different resolutions, which enables each pixel in the depth map to perceive more pixels in the source image and improves the performance of the model. As for the locality of photometric error, we propose a structural similarity (SSIM) pyramid loss to allow the model to sense the difference between images in multiple areas of different sizes. Experimental results show that our method achieves superior performance on both outdoor and indoor benchmarks.

Adaptive Multiscale Superpixel Embedding Convolutional Neural Network for Land Use Classification

Currently, large quantities of remote sensing images with different resolutions are available for earth observation and land monitoring, which are inevitably demanding intelligent analysis techniques for accurately identifying and classifying land use (LU). This article proposes an adaptive multi-scale superpixel embedding convolutional neural network architecture (AMUSE-CNN) for tackling land use classification. Initially, the images are parsed via the superpixel representation so that the object based analysis (via a superpixel embedding CNN scheme) can be carried out with the pixel context and neighborhood information. Then, a multi-scale convolutional neural network (MS-CNN) is proposed to classify the superpixel based images by identifying object features across a variety of scales simultaneously in which multiple window sizes are used to fit to the various geometries of different LU classes. Furthermore, a proposed adaptive strategy is applied to best exert the classification capability of MS-CNN. Subsequently two modules are developed to fully implement the AMUSE-CNN architecture. More specifically, Module I is to determine the most suitable classes for each window size (scale) by applying majority voting to a series of MS-CNNs. Module II carries out the classification of the classes identified in Module I for the given scale used in MS-CNN and therefore complete the LU classification of the entire classes. The proposed AMUSE-CNN architecture is both quantitatively and qualitatively validated using remote sensing data collected from two cities, Kano and Lagos in Nigeria due to the spatially complex land use distribution. Experimental results show the superior performance of our approach against several state-of-the-art techniques.

Experimental study of the adaptive gain feature for improved position-sensitive ion spectroscopy with Timepix2

In the present work, we study the Timepix2 pixels’ high energy response in the so-called adaptive gain mode. Therefore, Timepix2 with a 500 μm thick silicon sensor was irradiated with protons of energies in the range from 400 keV to 2 MeV and α-particles of 5.5 MeV from 241Am. A novel method was developed to determine the energy deposit in single pixels of particle imprints, which are spread out over a set of neighbor pixels (cluster). We show that each pixel is capable of measuring the deposited energy from 4 keV up to ∼3.2 MeV. Reconstructing the full energy content of the clusters, we found relative energy resolutions () better than 2.7% and better than 4% for proton and α-particle data, respectively. In a simple experiment with a 5.5 MeV α-particle source, we demonstrate that energy losses in thin (organic) specimen can be spatially resolved, mapping out sample thickness variations, with a resolution around 1–2 μm, across the sensor area. The inherent spatial resolution of the device was determined to be 350 nm in the best case.

Reconstruction of Connected Digital Lines Based on Constrained Regularization.

This paper presents a new approach for reconstruction of disconnected digital lines (DDLs) based on a constrained regularization model which ensures connectivity of the digital lines (DLs) in the discrete image plane. The first step in this approach is to determine the order of given pixels of the DDL. To determine connectivity of pixels, we use the usual 8-neighbor connectivity in discrete images. For any neighboring pixels of the DDL that are not connected, we determine a number of new pixel values that need to be reconstructed between these pixels. Next, the integer-valued x - and y -coordinates of the location of the pixels of the DDLs are segregated into two 1D signal vectors. Then the x - and y -coordinates of the missing pixels of the DDLs are estimated using a new constrained regularization. While the solution of this constrained minimization problem provides real values for the x - and y -coordinates of pixels positions, the imposed constraint ensures connectivity of the resulting DLs in the image plane after transforming the computed values from [Formula: see text] to [Formula: see text]. The proposed regularization approach forces connected lines with small curvature. The experimental results demonstrate that the proposed technique improves DL intersection detection, as well. Moreover, this technique has a high potential to be used as a fast approach in binary image inpainting particularly overcoming the shortcomings of conventional methods which cause destruction of thin objects and blurring in the recovered regions.

Recovering Realistic Details for Magnification-Arbitrary Image Super-Resolution.

The emergence of implicit neural representations (INR) has shown the potential to represent images in a continuous form by mapping pixel coordinates to RGB values. Recent work is capable of recovering arbitrary-resolution images from the continuous representations of the input low-resolution (LR) images. However, it can only super-resolve blurry images and lacks the ability to generate perceptual-pleasant details. In this paper, we propose implicit pixel flow (IPF) to model the coordinate dependency between the blurry INR distribution and the sharp real-world distribution. For each pixel near the blurry edges, IPF assigns offsets for the coordinates of the pixel so that the original RGB values can be replaced by the RGB values of a neighboring pixel which are more appropriate to form sharper edges. By modifying the relationship between the INR-domain coordinates and the image-domain pixels via IPF, we convert the original blurry INR distribution to a sharp one. Specifically, we adopt convolutional neural networks to extract continuous flow representations and employ multi-layer perceptrons to build the implicit function for calculating pixel flow. In addition, we propose a new double constraint module to search for more stable and optimal pixel flows during training. To the best of our knowledge, this is the first method to recover perceptually-pleasant details for magnification-arbitrary single image super-resolution. Experimental results on public benchmark datasets demonstrate that we successfully restore shape edges and satisfactory textures from continuous image representations.

Filtering and Enhancement Method of Ancient Architectural Decoration Image based on Neural Network

Due to poor ambient light or uneven lighting, the old decoration image acquisition methods are easy to cause the image blur. To solve this problem, this paper proposes a neural network-based filtering enhancement method for ancient architectural decoration images, which preserves image details by enhancing contrast, smoothing noise reduction and edge sharpening. Based on the convolutional neural network which is composed of encoder, decoder and layer hop connection, the residual network and hole convolution are introduced, and the hole U-Net neural network is constructed to fuse the pixel feature blocks of different levels. This method enhanced the image contrast according to the gray level and frequency histogram, and aiming at the gray value of the pixel to be processed in the image. And the middle value of the gray value of the neighborhood pixel is used to filter the noise of the ancient building decoration image. The paper also analyzes the joint strength of beams and columns in ancient buildings, and calculates the elastic constants of beams and columns and the stress at the joint of them, considering the image texture characteristics of the wood in ancient buildings with the mortise and tenon connection of beams and columns. Experimental results show that the proposed method has good noise suppression performance, can effectively obtain image detail features, and significantly improve the subjective visual effect of ancient architectural decoration images.

Adaptive Registration for Optical and SAR Images With a Scale-Constrained Matching Method

Images registration for optical and synthetic aperture radar (SAR) is a key of multi-sensor images analysis. And parameters selection affects the final result of the registration algorithms for optical and SAR images. For images obtained by different sensors, how to choose an appropriate parameter for accurate registration is a key problem. In this letter, a parameter adaptive registration algorithm between optical and SAR images based on SIFT is proposed. Because of the different imaging mechanisms of these two kinds of images, the algorithm uses the multiscale Sobel operator to calculate the gradient for the optical image, while for the SAR image, a new adaptive operator based on the neighborhood pixel value is proposed to calculate the gradient. In the feature extraction, the adaptive value estimated by constant false alarm rate (CFAR) detection is used instead of the fixed threshold. Finally, a matching method constrained by image size scaling (Scale-Constrained Fast Sample Consensus) is proposed. The evaluation was designed in two aspects: feature extraction and image registration. The algorithm we proposed shows excellent performances in the aspects of repeatability, correct matching rate and root mean square error. The experimental results show that our method maintains the performance of the original algorithm and has some optimization and breakthrough.

Classification Method for Power Load Data of New Energy Grid based on Improved OTSU Algorithm

The classification method for power load data of new energy grid based on improved OTSU algorithm is studied to improve the classification accuracy of power load data. According to the idea of two-dimensional visualization of time series, GAF (Geographical Adaptive Fidelity) method is used to transform the current data of power load in new energy grid into two-dimensional image of power load in new energy grid. The intra class dispersion is introduced, and the improved OTSU algorithm is used to segment the foreground and background of the two-dimensional image according to the pixel gray value of the two-dimensional image and the one-dimensional inter class variance corresponding to the pixel neighborhood gray value. The two-dimensional foreground image of power load is taken as the input sample of convolution neural network. The convolution neural network extracts the features of the two-dimensional foreground image of power load through convolution layer. According to the extracted features, the classification results of power load data of new energy grid are output through three steps: nonlinear processing, pooling processing and full connection layer classification. The experimental results show that this method can accurately classify the power load data of new energy grid, and the classification accuracy is higher than 97%.

Dynamic Graph-Level Neural Network for SAR Image Change Detection

The graph neural network (GNN) has been widely applied to image analysis and recognition. Recently, a semisupervised graph convolutional network (ssGCN) method has been proposed to change detection and obtains promising performance on very-high-resolution remote sensing images. However, a synthetic aperture radar (SAR) image is subject to speckle noise, and there is no explicit structure. In this letter, an end-to-end dynamic graph-level neural network (DGLNN) is proposed to exploit the local structure of each pixel neighborhood block at a graph level and learn a more discriminative graph for change detection. Moreover, in the training of DGLNN, a <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$K$ </tex-math></inline-formula> -nearest neighborhood is employed to reconstruct edges between nodes instead of the fixed edges between two nodes so that each node exploits the features from different neighbor nodes. The proposed method is verified by cross-domain SAR image change detection on four sets of SAR images and compared with five state-of-the-art deep-learning-based SAR image change detection methods. The overall experimental results show that the proposed DGLNN obtains outstanding performance.

An information technique for segmentation of military assets in conditions of uncertainty of initial data

To solve the applied task of detecting military assets in aerospace images the presented paper investigates the processes of constructing segmented maps of the images. The goal is to develop an information technique for detecting military assets in conditions of uncertainty of initial data. To achieve the goal, the following tasks were formulated: 1) to analyze usability of the existing segmentation methods for automatic detection of military assets in the images; 2) if the existing methods are inapplicable, to develop a new algorithm to solve the problem. In the paper the following methods are used: the methods of digital image processing, the methods of Boolean algebra and fuzzy sets, the methods of statistical analysis. The following results are received. Analysis of the known segmentation methods showed that due to camouflage coloring of the military assets, similarity of their color characteristics to those of underlying surfaces and due to the presence of large number of textured fragments in the images those methods provide segmented maps of poor quality. Among the common problems arising when conventional methods are used there are wrong segmentation, when the received contours do not coincide with the borders of the objects of interest; oversegmentation, when there are a lot of minor segments which produce "litter" objects; undersegmentation, when potentially possible segments are missed etc. As the conventional methods are inapplicable, in the paper it is suggested using the fuzzy logic systems. For each pixel the probability of the fact that the pixel belongs to the object or to the background is calculated. For making decision whether a pixel belongs to the object the production rules based on the chosen most significant factors (probabilistic values of spectral sub-bands, belonging of the neighboring pixels to the object, jumps of brightness in spectral sub-bands on the object's borders) are constructed. Conclusion. The suggested technique ensures high-quality definition of objects' borders, thus considerably increasing the reliability of military assets recognition.

A novel feature descriptor based coral image classification using extreme learning machine with ameliorated chimp optimization algorithm

Coral reefs are one of the most biodiverse and productive ecosystems on earth, which acts as a habitat and provides food to a wide range of organisms. The coral reefs protect the coasts from storms, waves, and erosion. Unfortunately, this ecosystem is under threat by various natural and manmade factors, including pollution, ocean acidification, sedimentation, and climate change. The marine ecosystems need to be monitored regularly in order to minimize the aforementioned threats. Hence, an automated classification system is necessary for the classification of coral images. In this paper, a novel method called “Systematic Local Directional Encoded Derivative Binary Pattern” (SLDEDBP) is proposed to extract the coral reef feature values. In this descriptor, the feature values are derived using two phases: Local Directional Derivative Binary Pattern (LDDBP) and Systematic Local Discriminate Binary pattern (SLDBP). In the LDDBP phase, the eight directional kirsch mask is employed in each 3 × 3 neighbourhood window to extract the directional edge response of an image. Then, in the SLDBP phase, the local neighbourhood pixel values are divided into four non-overlapping groups of Z, M, TZ (Tilted Z), and TM (Tilted M) to extract unique features values by considering the centre pixel value in a 3 × 3 neighbourhood window. Finally, the derived feature values of LDDBP and SLDBP are concatenated to procedure final discriminate feature values. The extracted feature values of coral images are applied to a hybridized method called Ameliorated Chimp Optimization Algorithm with Extreme Learning Machine Classifier (AChOA-ELM) to optimize the relative weight of the input and biases of neural networks. This avoids overfitting and makes the model more generalized and robust. The proposed AChOA-ELM method performance is evaluated using different databases of coral and standard images. The reliability of the proposed model has been evaluated utilizing the model performance metrics, accuracy (ACC), true positive rate (TPR), true negative rate (TNR), false-positive rate (FPR), precision (PRC), positive predictive value (PPV), and F1 score. The findings reveal that the proposed hybrid AChOA-ELM achieves greater classification accuracy with a faster coverage rate than other existing methods. In the presence of the image data augmentation technique, the EILAT2 dataset achieves the highest classification accuracy of 99%, whereas MLC dataset achieves the lowest classification accuracy of 94.86%.

Ensemble solar forecasting and post-processing using dropout neural network and information from neighboring satellite pixels

Ensemble weather forecasts are often found to be under-dispersed and biased. Post-processing using spatio-temporal information is, therefore, required if one wishes to improve the quality of the raw forecasts. It is on this account that the present article generates and post-processes ensemble solar forecasts using satellite-derived irradiance not only from the focal pixel but also from the neighboring pixels. The ensemble forecasting model of choice is a dropout neural network with Monte Carlo sampling, eliminating the need for training multiple models and ensuring parameter diversity in ensemble forecasting. Subsequently, ensemble forecasts are post-processed using both parametric and nonparametric post-processing techniques, such as nonhomogenous regression, generalized additive model, linear quantile regression, or quantile random forests. The proposed forecasting framework is demonstrated and verified using four years of half-hourly data, at seven locations in the United States. Continuous ranked probability skill scores as high as 66% have been obtained when comparing the proposed method to a conditional climatology reference. The content of this article may be useful to a wide range of stakeholders in the power system, including but not limited to: independent system operators, who aim at efficiently maintaining the system’s reliability; utility- and distributed-scale PV plant owners, who wish to avoid penalties for power deviation between the scheduled and real-time delivery; and forecast retailers, who can benefit from selling solar forecasts of higher quality.

Adaptive neighbor constrained deviation sparse variant fuzzy c-means clustering for brain MRI of AD subject

Progression of Alzheimer’s disease (AD) bears close proximity with the tissue loss in the medial temporal lobe (MTL) and enlargement of lateral ventricle (LV). The early stage of AD, mild cognitive impairment (MCI), can be traced by diagnosing brain MRI scans with advanced fuzzy c-means clustering algorithm that helps to take an appropriate intervention. In this paper, firstly the sparsity is initiated in clustering method that too rician noise is also incorporated for brain MR scans of AD subject. Secondly, a novel neighbor pixel constrained fuzzy c-means clustering algorithm is designed where topoloty-based selection of parsimonious neighbor pixels is automated. The adaptability in choice of neighbor pixel class outliers more justified object edge boundary which outperforms a dynamic cluster output. The proposed adaptive neighbor constrained deviation sparse variant fuzzy c-means clustering (AN_DsFCM) can withhold imposed sparsity and withstands rician noise at imposed sparse environment. This novel algorithm is applied for MRI of AD subjects and normative data is acquired to analyse clustering accuracy. The data processing pipeline of theoretically plausible proposition is elaborated in detail. The experimental results are compared with state-of-the-art fuzzy clustering methods for test MRI scans. Visual evaluation and statistical measures are studied to meet both image processing and clinical neurophysiology standards. Overall the performance of proposed AN_DsFCM is significantly better than other methods.

Using Median-LBPH Algorithm for Real-Time Face Recognition System

Abstract: The LBPH algorithm is used ubiquitously for Face Recognition applications in modern times because of its simplicity of implementation, despite providing high accuracy and less computation time. However, in conditions of varied illumination, face expression and angles at which face images are captured, its confidence is decreased. We propose a slightly modified algorithm that considers the median of the neighbourhood pixels rather than the pixel itself to overcome this issue. This algorithm is called Median-LBPH. The grey value of every pixel is replaced by the median of all the neighbourhood pixel values. Then the features are extracted, and a histogram representing the original image is saved in the model. This model, in turn, can be used to compare with histograms obtained from the faces in real-time footage to find a potential match. This algorithm is used in an end-to-end face recognition system, a web application prototype for Law Enforcement Agencies to maintain a central criminal database shared and accessed across various departments. A live surveillance system is added as part of this novel application so that whenever an already registered criminal appears live on surveillance cameras, a notification will be received, and personnel appropriate Law Enforcement authorities will receive e-mail and text messages through a secured channel. Keywords: Face Recognition, Median-Local Binary Pattern Histogram (MLBPH), Haar Cascade, Adaboost, Neighbourhood Median

Polarimetric SAR Speckle Filtering Using a Nonlocal Weighted LMMSE Filter.

Despeckling is a key preprocessing step for applications using PolSAR data in most cases. In this paper, a technique based on a nonlocal weighted linear minimum mean-squared error (NWLMMSE) filter is proposed for polarimetric synthetic aperture radar (PolSAR) speckle filtering. In the process of filtering a pixel by the LMMSE estimator, the idea of nonlocal means is employed to evaluate the weights of the samples in the estimator, based on the statistical equalities between the neighborhoods of the sample pixels and the processed pixel. The NWLMMSE estimator is then derived. In the preliminary processing, an effective step is taken to preclassify the pixels, aiming at preserving point targets and considering the similarity of the scattering mechanisms between pixels in the subsequent filter. A simulated image and two real-world PolSAR images are used for illustration, and the experiments show that this filter is effective in speckle reduction, while effectively preserving strong point targets, edges, and the polarimetric scattering mechanism.

A machine learning-based approach for picture acquisition timeslot prediction using defective pixels

Estimating the acquisition time of digital photographs is a challenging task in temporal image forensics, but the application is highly demanded for establishing temporal order among individual pieces of evidence and deduce the causal relationship of events in a court case. The forensic investigator needs to identify the timeline of events and look for some patterns to gain a clear overview of activities associated with a crime. This paper aims to explore the presence of defective pixels over time for estimating the acquisition date of digital pictures. We propose a technique to predict the acquisition timeslots of digital pictures using a set of candidate defective pixels in non-overlapping image blocks. First, potential candidate defective pixels are determined through related pixel neighbourhood and two proposed features, called the local variation features to best fit in a machine learning model. The machine learning approach is used to model the temporal behaviour of camera sensor defects in each block using the scores obtained from individually trained pixel defect locations and fused in a majority voting method. Interestingly, timeslot estimation using individual blocks has been shown to be more accurate when virtual sub-classes corresponding to halved timeslots are first considered prior to the reconstruction step. Finally, the last stage of the system consists of the combination of block scores in a second majority voting operation to further enhance performance. Assessed on the NTIF image dataset, the proposed system has been shown to reach very promising results with an estimated accuracy between 88% and 93% and clear superiority over a related state-of-the-art system.