Central Pixel Research Articles

A Completed Multiply Threshold Encoding Pattern for Texture Classification

The binary pattern family has drawn wide attention for texture representation due to its promising performance and simple operation. However, most binary pattern methods focus on local neighborhoods but ignore center pixels. Even if some studies introduce the center based sub-pattern to provide complementary information, extant center based sub-patterns are much weaker than other local neighborhood based sub-patterns. This severe unbalance limits the classification performance of fusion features significantly. To alleviate this problem, this paper designs a multiply threshold center pattern (MTCP) to provide a more discriminative and complementary local texture representation with a compact form. First, a multiply thresholds encoding strategy is designed to encode the center pixel that generates three 1-bit binary patterns. Second, it adopts a compact multi-pattern encoding strategy to combine them into the 3-bit MTCP. Furthermore, this paper proposes a completed multiply threshold encoding pattern by fusing the MTCP, local sign pattern, and local magnitude pattern. Comprehensive experimental evaluations on three popular texture classification benchmarks confirm that the completed multiply threshold encoding pattern achieves superior texture classification performance.

Зависимость величины сигнала от смещения кружка рассеяния объектива в диапазоне углов 0º - 90º относительно центра пиксела фоточувствительной матрицы

An algorithm has been developed for calculating the change in the irradiation level of a matrix pixel when the focal spot shifts relative to the center of the matrix pixel in the range from 0 ° to 90 °. Two types of pixel irradiance distribution are considered: distribution in the lens scattering circle in the form of a gaussian rotation and uniform irradiance. Method. The calculation is based on the method of dividing the lens scattering circle on the matrix pixel into separate areas, according to which the irradiance is calculated. The displacement of the lens scattering circle by ∆x along the X axis and by ∆y along the Y axis is taken as normalized to the spot radius. To create a two-dimensional graph of the dependence of the pixel irradiance on the displacement of the lens scattering circle along the X, Y axes, we introduce the transformed relative displacement of the lens scattering circle γ. The main results. An algorithm has been developed for calculating the change in the irradiation level of a matrix pixel when the lens scattering circle is shifted relative to the matrix pixel for two cases of pixel irradiation: the distribution in the lens scattering circle in the form of a gaussian rotation and uniform irradiation. The dependence of the normalized irradiance of the matrix pixel on the transformed relative displacement of the lens scattering circle γ at an angle of 0 °, 30 °, 45 °, 60 °, 90 ° is plotted on the graph for the above cases of irradiation.

CNN and Transformer interaction network for hyperspectral image classification

ABSTRACT Convolutional Neural Network (CNN) has developed hyperspectral image (HSI) classification effectively. Although many CNN-based models can extract local features in HSI, it is difficult for them to extract global features. With its ability to capture long-range dependencies, Transformer is gradually gaining prominence in HSI classification, but it may overlook some local details when extracting features. To address these issues, we proposed a CNN and transformer interaction network (CTIN) for HSI classification. Firstly, A dual-branch structure was constructed in which CNN and Transformer are arranged in parallel to simultaneously extract global features and local features in HSI. Secondly, a feature interaction module has been imported between the two branches, thus facilitating a bi-directional flow of information between the global and local feature spaces. In this way, the network structure combines the advantages of CNN and Transformer in extracting features as much as possible. In addition, a token generation method is designed to harness abundant contextual information that is relevant to the centre pixel, and improve the accuracy of the final classification. Experiments were conducted on four hyperspectral datasets (two classical datasets – Indian Pines, Salinas Valley, a new satellite dataset – Yellow River, and an self-made UAV dataset-Yellow River Willow). Experimental results show that the proposed method outperforms the other state-of-the-art methods, with overall accuracies of 99.21%, 99.61%, 92.40%, and 98.17%, respectively.

Noise-suppression algorithm of GM-APD lidar based on Markov random field

Background light easily affects a Geiger-mode avalanche photodiode (GM-APD) laser imaging, detection, and ranging (Lidar), and its detection ability is significantly reduced in strong light environments. Improving imaging performance under a low signal-to-back ratio has become critical. A GM-APD lidar noise-suppression algorithm based on a Markov random field is proposed. The field objective function and prior model are established using similar centre and adjacent pixel features. Both enhance the feature difference between a strong background light and a weak echo signal and improve the ability of weak echo signal reconstruction under an ultra-low signal-to-background ratio (SBR). Imaging experiments of long-distance building targets under varying illumination verified the noise-suppression ability of the algorithm. When the SBR is 0.0055, target recovery can reach 19.40%, and recovery of the whole array image can reach 60.94%. Compared with the sparse Poisson intensity reconstruction algorithm, the image signal-to-noise ratio is improved by 83 dB, and the average depth error was reduced by 36.14 cm. The proposed algorithm improves the imaging performance under ultra-low SBR; it significantly promotes the development of GM-APD lidar all-time applications.

SEGMENTATION AND IDENTIFICATION OF DISEASES IN RETINAL BLOOD VESSEL USING WIDE AND DEEP NEURAL NETWORK SYSTEM

A new supervised method for vessel segmentation in retinal images is studied here. Proposed method remolds the task of segmentation as a problem of cross-modality data transformation from retinal image to vessel map. A wide and deep neural network with strong induction ability is proposed to model the transformation, and an efficient training strategy is presented. Instead of a single label of the center pixel, the network can output the label map of all pixels for a given image patch. Our approach outperforms reported state-of-the-art methods in terms of sensitivity, specificity and accuracy. The result of cross-training evaluation indicates its robustness to the training set. The approach needs no artificially designed feature and no preprocessing step, reducing the impact of subjective factors. The proposed method has the potential for application in image diagnosis of ophthalmologic diseases, and it may provide a new, general, high- performance computing framework for image segmentation. Key Words- new supervised method, cross-modality, cross-training evaluation.

A Method for Extracting a Laser Center Line Based on an Improved Grayscale Center of Gravity Method: Application on the 3D Reconstruction of Battery Film Defects

Extraction of the laser fringe center line is a key step in the 3D reconstruction of linear structured light, the accuracy of which is directly related to the quality of the 3D model. A laser center line extraction method based on an improved gray center of gravity method is proposed to solve the problem of low extraction accuracy. Firstly, a smoothing method is used to eliminate the flat top of the laser line, and the Gaussian curve is adopted to fit the peak position of the curve. Then, the gray threshold is set to automatically extract the laser linewidth, and based on the window opening, the grayscale center of gravity method is improved to extract the coordinates of the center pixel for the second time. Finally, experiments show that the average absolute error of the improved laser line extraction method is 0.026 pixels, which is 2.3 times lower than the gray center of gravity method, 1.9 times lower than the curve fitting method, and the standard error can reach 0.005 pixels. Compared with the gray center of gravity method and the curve fitting method, the influence of gray value change on the center line extraction is more fully considered, and the center of the light strip can be extracted more accurately, achieving sub-pixel accuracy.

Extended right-angle difference ternary co-relation pattern: A new feature descriptor for face anti-spoofing

Presently, in many applications face recognition is commonly used as an authentication security system. However, an intruder can spoof the face recognition system by using genuine user photos or video to gain access. Therefore, there is a need for a superior technique for face anti-spoofing to handle all kinds of spoofing attacks in any environment. This paper presents a novel descriptor called extended right-angle difference ternary co-relation pattern (ERDTCP) for face anti-spoofing. The ERDTCP is calculated based on the division of a given image into 4 blocks at an angle of 0°, 90°, 180° and 270° from the center pixel. The effectiveness of the proposed method is evaluated on three publically available databases using KNN and SVM classifiers. Experiments were conducted using SOTA approaches and the proposed method using ERDTCP and the results were compared. The proposed face anti-spoofing approach performs better than SOTA approaches.

A Multi-Scale Progressive Collaborative Attention Network for Remote Sensing Fusion Classification.

With the development of remote sensing technology, panchromatic images (PANs) and multispectral images (MSs) can be easily obtained. PAN has higher spatial resolution, while MS has more spectral information. So how to use the two kinds of images' characteristics to design a network has become a hot research field. In this article, a multi-scale progressive collaborative attention network (MPCA-Net) is proposed for PAN and MS's fusion classification. Compared to the traditional multi-scale convolution operations, we adopt an adaptive dilation rate selection strategy (ADR-SS) to adaptively select the dilation rate to deal with the problem of category area's excessive scale differences. For the traditional pixel-by-pixel sliding window sampling strategy, the patches which are generated by adjacent pixels but belonging to different categories contain a considerable overlap of information. So we change original sampling strategy and propose a center pixel migration (CPM) strategy. It migrates the center pixel to the most similar position of the neighborhood information for classification, which reduces network confusion and increases its stability. Moreover, due to the different spatial and spectral characteristics of PAN and MS, the same network structure for the two branches ignores their respective advantages. For a certain branch, as the network deepens, characteristic has different representations in different stages, so using the same module in multiple feature extraction stages is inappropriate. Thus we carefully design different modules for each feature extraction stage of the two branches. Between the two branches, because the strong mapping methods of directly cascading their features are too rough, we design collaborative progressive fusion modules to eliminate the differences. The experimental results verify that our proposed method can achieve competitive performance.

Modulation depth of the gyrosynchrotron emission as an identifier of fundamental sausage modes

Aims. We aim to study the intensity, the modulation depth, and the mean modulation depth of the gyrosynchrotron (GS) radiation as a function of the frequency and the line of sight (LOS) in fast sausage modes. Methods. By solving the 2.5D magnetohydrodynamics (MHD) ideal equations of a straight coronal loop considering the chromosphere and with typical flaring plasma parameters we analyse the wavelet transform of the density and the GS emission for different radio frequencies and different spatial resolutions, given impulsive and general perturbations with energies in the microflare range. Results. A wavelet analysis performed over the GS radiation emission showed that a fast fundamental sausage mode of ∼7 s with a first harmonic mode of 3 s developed, for all the initial energy perturbations used. For both the high spatial resolution (central pixel integration) and the low spatial resolution (entire loop integration), the larger the radio frequency, the larger the modulation depth. However, high- and low-resolution integrations differ in that the larger the LOS angle with respect to the loop axis, the larger and smaller the modulation depth, respectively. Conclusions. Fast MHD modes triggered by instantaneous energy depositions of the order of a microflare energy are able to reproduce deep intensity modulation depths in radio emission as observed in solar events. As the trends of the GS emission previously obtained for a linear and forced oscillation remain present when analysing a more general context, considering the chromosphere and where the sausage mode is triggered by an impulsive, non-linear perturbation, it seems that the behaviour found can be used as observational identifiers of the presence of sausage modes with respect to other quasi-periodic pulsation features. It can be inferred from this that finite-amplitude sausage modes have the potential to generate the observed deep modulation depths.

3D CNN and grad-CAM based visualization for predicting generation of dislocation clusters in multicrystalline silicon

We propose a machine learning-based technique to address the crystallographic characteristics responsible for the generation of crystal defects. A convolutional neural network was trained with pairs of optical images that display the characteristics of the crystal and photoluminescence images that show the distributions of crystal defects. The model was trained to predict the existence of crystal defects at the center pixel of the given image from its optical features. Prediction accuracy and separability were enhanced by feeding three-dimensional data and data augmentation. The prediction was successful with a high area under the curve of over 0.9 in a receiver operating characteristic curve. Likelihood maps showing the distributions of the predicted defects are in good resemblance with the correct distributions. Using the trained model, we visualized the most important regions to the predicted class by gradient-based class activation mapping. The extracted regions were found to contain mostly particular grains where the grain boundaries changed greatly due to crystal growth and clusters of small grains. This technique is beneficial in providing a rapid and statistical analysis of various crystal characteristics because the features of optical images are often complex and difficult to interpret. The interpretations can help us understand the physics of crystal growth and the effects of crystallographic characteristics on the generation of detrimental defects. We believe that this technique will contribute to the development of a better fabrication process for high-performance multicrystalline materials.

Impulse Noise Suppression in Color Images Using Median Filter and Deep Learning

Abstract: Refining the quality of a noisy image is essential for many image applications. Various median filter variants have been introduced to suppress various noises, but they have their own limitations when detecting noise and restoring noise-free images. Denoising convolutional neural networks (DnCNNs), primarily developed for Gaussian noise removal, are influential nonlinear mapping models in image processing. After alterations in training data, they can be used to suppress other noise with outstanding results. This article suggests a frequency median filter method combined with deep learning for color images polluted by Salt and Pepper (SnP) noise. The analysis presented in this paper has primarily used a frequency median filter to suppress impulse noise wherein the restored value for the center pixel is evaluated by the frequency median rather than the traditional median. After which, the pretrained denoising convolutional neural network is hired to suppress the remaining noise and attain the output image finally. With a visual comparative study, simulation results on the taken test images show that the proposed method surpasses de-noising methods in terms of PSNR, SSIM, NMSE, Entropy, IEF, NCC, PCC and Running Time.

기울기 기반 허프 변환을 이용한 실시간 차선 검출 시스템 설계

Straight line detection using the gradient-based Hough transform has high accuracy, but has the disadvantage of increased memory and longer processing time due to storing and processing a lot of data. The proposed architecture applies the gradient-based Hough transform algorithm to consider only the points in the range of -4 to +4 centered on the gradient angle in parallel, and increases accuracy by applying the highest weight to the central pixel. In order to improve overall performance, the LUT for the trigonometric functions and voting memory for weight values are divided into 9 parts for each address area, so that the line detection speed is improved by processing data at the same time. The proposed architecture was designed with VerilogHDL and synthesized with DBhitek 110nm standard cell library and the results showed that the maximum operating frequency was 210Mhz (73 fps) with 15K of 2-input NAND gates and 5,950Kb of memory.

Dual-View Hyperspectral Anomaly Detection via Spatial Consistency and Spectral Unmixing

Anomaly detection is a crucial task for hyperspectral image processing. Most popular methods detect anomalies at the pixel level, while a few algorithms for anomaly detection only utilize subpixel level unmixing technology to extract features without fundamentally analyzing the anomalies. To better detect and separate the anomalies from the background, this paper proposes a dual-view hyperspectral anomaly detection method by taking account of the anomaly analysis at both levels mentioned. At the pixel level, the spectral angular distance is adopted to calculate the similarities between the central pixel and its neighbors in order to further mine the spatial consistency for anomaly detection. On the other hand, from the aspect of the subpixel level analysis, it is considered that the difference between the anomaly and the background usually arises from dissimilar endmembers, where the unmixing will be fully implemented. Finally, the detection results of both views are fused to obtain the anomalies. Overall, the proposed algorithm not only interprets and analyzes the anomalies from dual levels, but also fully employs the unmixing for anomaly detection. Additionally, the performance of multiple data sets also confirmed the effectiveness of the proposed algorithm.

Learning pixel-adaptive weights for portrait photo retouching

The lookup table-based methods achieve promising retouching performance by learning image-adaptive weights to combine 3-dimensional lookup tables (3D LUTs) and conducting pixel-to-pixel color transformation. However, this paradigm ignores the local context cues and applies the same transformation to portrait pixels and background pixels that exhibit the same raw RGB values. In contrast, an expert usually conducts different operations to adjust the color temperatures, tones of portrait regions, and background regions. This inspires us to model local context cues to improve the retouching quality explicitly.Thus, the center pixel of an image patch is first retouched by predicting pixel-adaptive lookup table weights. To modulate the influence of neighboring pixels, as neighboring pixels exhibit different affinities to the center pixel, a local attention mask is estimated. Then, the quality of the local attention mask is further improved by applying supervision, which is based on the affinity map calculated by the ground-truth portrait mask. For group-level consistency, we propose to directly constrain the variance of mean color components in the Lab space. Extensive experiments on the PPR10K dataset demonstrate the effectiveness of the proposed method, the retouching performance on high-resolution photos is improved by over 0.5dB in terms of PSNR, and the group-level inconsistency is reduced by 2.1.

Image Splicing Forgery Detection using Standard Division-Local Binary Pattern Features

Numerous aspects of daily life contribute to societal stability, and the security of people's perceptions of the world online is one target of various malicious attacks. Professional forgers can now quickly create copy-move, splice, or retouch photos with the use of today's advanced tools. It has been determined that splicing, is a widespread method of manipulating images. Image forgery can also lead to substantial setbacks and challenges, some of which may have significant ethical, moral, or legal consequences. Thus, the paper proposes a system that combines SD-LBP (Standard Devision-Local Binary Pattern) based passive picture splicing detection system and ANN classifier. The SD-LBP is created to have the benefits and avoid drawbacks of Local Binary Pattern (LBP). The SD-LBP extraction is typically performed by employing proposed SD value-based thresholding instead of the center pixel, which is robust to noise and other photometric attacks. The second part of the proposed system is the ANN classifier is used that extract the feature of images to lower the error and build a model that can tell spliced images from real photos that have been digitally altered. The proposed system is creating a reliable image forgery detection technique that was implemented with CASIA V2.0 standard dataset. The results showing that it outperformed compared with other methods on the in terms of accuracy (97.8%), sensitivity (98.6%), and specificity (97.1%). Most importantly, the proposed SFD method exceeded the state-of-the-art efforts in this field in terms of accuracy.

Completed homogeneous LBP for remote sensing image classification

ABSTRACT Land cover (LC) classification remains a challenging task due to the diversity of terrain and topography, limited prior knowledge, and complex data sets. These variations can lead to differences in illumination and shading, which can affect the appearance of objects in the image. The local binary pattern (LBP) model is an effective technique for capturing local texture information in an image, which can help to overcome the effects of topography diversity by analysing patterns in different image regions, even if the illumination or shading conditions are different. However, LBP alone is inadequate for characterizing high-resolution remotely sensed images with complex semantic content as it only utilizes sign information in the local region. In this paper, a new texture characterization descriptor, known as completed homogeneous LBP (CHLBP), is proposed as an improved version of homogeneous LBP (HLBP) for LC classification of remotely sensed images. The CHLBP method mainly involves the following steps: first, sign and magnitude information from the HLBP descriptor is extracted, providing an effective alternative to the HLBP complementary contrast measure. Second, for each sign and magnitude function, a new splitting factor is used to obtain a depth relationship between the centre and its neighbouring pixels and to enhance noise robustness. Finally, the centre pixels representing the image grey level are also considered to contain discriminative information. The performance of our descriptor is evaluated using four challenging texture databases: Outex (TC10, TC12), Geofen Image Dataset (GID), and large-scale aerial (AID). Extensive experiments were performed using four classifiers: Support Vector Machine (SVM), K-Nearest Neighbours (KNN), Random Forest (RF), and Multi-Layer Perceptron (MLP), demonstrating the effectiveness and robustness of our descriptor against noise and free noise conditions in public remote sensing datasets.

Multi-scale image depth fusion method based on superpixel-level convolutional neural network

In order to improve the image quality, reduce the image noise and improve the image definition, the image depth fusion processing is realized by using the sp CNN network (Super pixel level convolution neural network, sp CNN). The improved non-local mean method is used to de-noise the image to highlight the role of the center pixel of the image block; the de-noised image is segmented by the improved CV model (Chan-Vese, CV), and the globally optimal multi-scale image segmentation result is obtained after optimization; From the perspective of regional features, the similarity measurement of image regions is carried out to realize image preprocessing. The sp-CNN network is constructed, and with the help of the idea of pyramid pooling, the average pooling is used to extract the features of each layer from the global and local levels of the convolutional features, and the training data set is generated for training, thereby realizing multi-scale image fusion. The experimental results show that the optimal value of the root mean square error index of the proposed method is 0.58. The optimal value of structural similarity index is 41.22. On the average slope index, the optimal value is 21.39. The optimal value of cross entropy index is 2.21. This shows that the proposed method has high image definition and good visual effect, which verifies the effectiveness of the method.

A reversible data hiding scheme for sharing medical data

In this paper, we propose a data hiding technique that helps the receiver to extract the secret data and recover the original image. The receiver first creates a pair of keys using Paillier cryptosystem and shares public key with sender. The sender enlarges cover image of size M × N to (2M − 1) × (2N − 1) by inserting one row (column) between two rows (columns), and then divides it into 3 × 3 blocks such that two boundary blocks have a row or a column in common, and two interior blocks have one row and one column in common. The block’s corner pixels, called as seed pixels, belong cover image and others, called as non-seed pixels, are initialized as 1 that are updated while embedding secret data in seed pixels. The block’s central pixels hide a secret bit using the parity and seed pixels are processed clockwise to hide two secret bits in each using separate procedures for even and odd pixels. The seed’s next pixel is multiplied by a fixed number that helps receiver to determine how it was processed at sender side. All pixels are encrypted using Paillier public key, giving a stegoimage, that is transmitted to the receiver. The receiver decrypts the stegoimage using Paillier private key, extracts the secret data and cover image by performing the data extraction process. The embedding rate of proposed method is 2.24 bits per pixel, which is 12 %, 16 %, 89 %, 83 %, 114 %, better than the Bhardwaj, Wang et al., Chen-Chang, Chi et al., Liu et al., respectively.

C58 A NEW MAGNETIC RESONANCE IMAGING TECHNIQUE TO ASSESS PROGNOSIS IN PATIENTS WITH HYPERTROPHIC CARDIOMYOPATHY AND LOW–INTERMEDIATE 5–YEAR RISK OF SUDDEN DEATH

Abstract Background Late gadolinium enhancement (LGE) is an important prognostic marker in hypertrophic cardiomyopathy and an extent >15% it is associated with high risk of sudden cardiac death. We proposed a novel method, the LGE–dispersion mapping, to assess heterogeneity of scar, and evaluated its prognostic role in patients with hypertrophic cardiomyopathy. Methods One hundred eighty–three patients with hypertrophic cardiomyopathy and a low– or intermediate 5–year risk of sudden cardiac death underwent cardiac magnetic resonance imaging. A parametric map was generated from each LGE image. A score from 0 to 8 was assigned at every pixel of these maps, indicating the number of the surrounding pixels having different quality (nonenhancement, mild–enhancement, or hyperenhancement) from the central pixel. The Global Dispersion Score (GDS) was calculated as the average score of all the pixels of the images. Results During a median follow–up time of 6 (25th–75th, 4–10) years, 22 patients had hard cardiac events (sudden cardiac death, appropriate implantable cardioverter–defibrillator therapy, resuscitated cardiac arrest, and sustained ventricular tachycardia). Kaplan–Meier analysis showed that patients with GDS>0.86 had worse prognosis than those with lower GDS (P<0.0001). GDS>0.86 was the only independent predictor of cardiac events (hazard ratio, 9.9 [95% CI, 2.9–34.6], P=0.0003). When compared with LGE extent >15%, GDS improved the classification of risk in these patients (net reclassification improvement, 0.39 [95% CI, 0.11–0.72], P<0.019). Conclusions LGE–dispersion mapping is a marker of scar heterogeneity and provides a better risk stratification than LGE presence and its extent in patients with hypertrophic cardiomyopathy and a low intermediate 5–year risk of sudden cardiac death.

Two-stage reverse knowledge distillation incorporated and Self-Supervised Masking strategy for industrial anomaly detection

In recent years, unsupervised anomaly detection based on knowledge distillation has gained special attention and some promising results have been reported in the literature. However, there is still room to improve the sensitivity of the model to anomalies. To do so, in this paper, a novel two-stage training method in terms of reverse knowledge distillation is proposed for anomaly detection and localization. Firstly, self-supervised mask training is introduced after the initial training of reverse knowledge distillation, which contributes greatly to the model detection against random unknown anomalies by self-simulating anomalies and forcing repair so as to reinforce learning single-category prototype patterns. Then, with the aim to facilitate the anomaly localization, an anomaly feature diffusion module is employed, which strengthens the correlation between pixels and helps spread the anomaly information to the surrounding area by covering the central pixel and reconstructing the representation for features after diffused. Furthermore, inspired by the human memory mechanism, an innovative normalized embedding memory bank is adopted to regulate the low-dimensional representations after embedding the encoding, inhibit the flow of anomalous information to the student decoder, and encourage the high-quality reconstruction of the model. Finally, the contextual similarity loss is used to guide the student model to learn knowledge representations from a contextual perspective, capture higher-order similarities between teachers and students, and delicately evaluate the differences between teachers and students. The empirical experiments conducted on the MVTec dataset show that the proposed SSMRKD method can achieve the best performance compared to other state-of-the-art methods, meanwhile extensive experiments of the ablation study validate the contribution of each component of the model. In addition, the advanced performance achieved on four commonly used datasets verifies the generalizability of the model in the industrial domain. Overall, the proposed SSMRKD method has significant advantages over the state-of-the-art anomaly detection methods.