International Journal of Image, Graphics and Signal Processing @ijigsp
Статьи журнала - International Journal of Image, Graphics and Signal Processing
Все статьи: 1168
Multi-Metric Based Face Identification with Multi Configuration LBP Descriptor
Статья научная
This paper deals with the performance improvement of a mono modal face identification. A statistical study of various structures of the LBPs (Local Binary Patterns) features associated to two metrics is performed to find out those committing errors on different subjects. Then, during the identification stage, these optimal variants are used, and a simple score level fusion is adopted. The score fusion is done after min-max normalization. The main contribution of this paper consists in the association of multiple LBP schemes with different metrics using simple fusion operation. The overall identification rating up to 99% using AT&T database is achieved.
Бесплатно
Multi-Module Convolutional Neural Network Based Optimal Face Recognition with Minibatch Optimization
Статья научная
Technology is getting smarter day by day and facilitating every part of human life from automatic alarming, automatic temperature, and personalised choice prediction and behaviour recognition. Such technological advancements are using different machine learning techniques for artificial intelligence. Face recognition is also one of the techniques to develop futuristic artificial intelligence-based technology used to get devices equipped with personalised features and security. Face recognition is also used for keeping information of facial data of employees of any company citizens of any country to get tracked and control over crimes in unfair incidents. For making face recognition more reliable and faster, several techniques are evolving every day. One of the fastest and most dependable face recognitions is CNN based face recognition. This work is designed based on the multiple convolutional module-based CNN equipped with batch normalisation and linear rectified unit for normalising and optimising features with minibatch. Faces in CNN’s fully connected layer are classified using the SoftMax classifier. The ORL and Yale face datasets are used for training. The average accuracy achieved is 94.74% for ORL and 96.60% for Yale Datasets. The convolutional neural network training was done for different training percentages, e.g., 66%, 67%, 68%, 69%, 70%, and 80%. The experimental outcomes exhibited that the defined approach had enhanced the face recognition performance.
Бесплатно
Multi-Scale and Auxiliary-Supervised Architectures for Accurate Road Network Mapping
Статья научная
Automated road network extraction from satellite imagery represents a critical advancement for Geographic Information Systems (GIS) applications in infrastructure management and urban planning. This paper introduces two novel deep learning architectures based on LinkNet: RoadNet-MS (Multi-Scale) and RoadNet-AUX (Multi-Scale with Auxiliary Supervision), specifically designed to enhance road segmentation performance. RoadNet-MS incorporates Multi-Scale Contextual Blocks (CMS-Blocks) and hybrid blocks to effectively capture diverse contextual features at multiple scales, achieving F1-scores of 78.87% on the challenging DeepGlobe dataset and 82.30% on the Boston & Los Angeles dataset. RoadNet-AUX extends this architecture through auxiliary supervision, further improving performance with F1-scores of 79.14% on DeepGlobe and 82.33% on Boston-LA. Both proposed architectures demonstrate competitive performance and consistent improvements over existing methods, including the state-of-the-art NL-LinkNet, across both evaluation datasets. Notably, RoadNet-MS achieves the highest precision (83.55%) among all compared methods on DeepGlobe. These contributions provide a pathway toward more accurate and scalable road network mapping, essential for modern urban planning and infrastructure monitoring applications.
Бесплатно
Multi-Stage Medical Image Encryption System Combining RSA and Steganography
Статья научная
Data security has become a major concern in the present era of the communication revolution, especially maintaining the confidentiality of medical images a prime concern in e-health establishments. As conventional techniques hold numerous drawbacks, this study aims to develop an image encryption algorithm by combining two renowned methods: the RSA algorithm and steganography. The proposed algorithm is modified with the help of the conventional RSA algorithm and steganography to provide an attainable solution to this alarming issue. RSA technique encrypts multiple medical images with distinct keys; further, these keys are embedded in two images to be transmitted secretly with the help of LSB steganography. The proposed system generates images of an unidentifiable pattern after encryption and decrypts those images without any loss. The claimed performances and robustness of the system are justified using different numerical and graphical measures such as PSNR, MSE, SSIM, NPCR, UACI, and histograms. This encryption method can be used for medical image transmission where image security is a vital concern.
Бесплатно
MultiBiometric Fusion: Left and Right Irises based Authentication Technique
Статья научная
Biometric science is one of the important applications in the pattern recognition field. There are several modalities used in the biometric applications, among these different traits we choose the iris modality. Therefore, this paper proposes a multi-biometric technique which combines the both units of the iris modality: the left and the right irises. The fusion combines the advantages of the two instances. For the both units of the iris, the segmentation is realized by a modified method and the feature extraction is done by a global approach (the Daubechies wavelets). The Support Vector Machine SVM is used to obtain scores for fusion. Then the scores obtained are normalized by Min-Max method and the fusion is performed at score level by the combination of two methods: a combination method with a classification method. The Fusion is tested using four databases which are: CASIAV4 database, SDUMLA-HMT database, MMU1, and MMU2 databases. The obtained results have confirmed that the multi-biometric systems are better than the mono-modal systems according to their performance.
Бесплатно
Статья научная
The fundamental component of the work contains a summary of the theoretical foundations of the algorithms of the scale-self-similar approach for the analysis of digital Mueller-matrix images of birefringent architectonics of biological tissues. The theoretical consideration of multifractal analysis and determination of singularity spectra of fractal dimensions of coordinate distributions of matrix elements (Mueller-matrix images - MMI) of biological tissue preparations is based on the method of maxima of amplitude modules of the wavelet transform (WTMM). The applied part of the work is devoted to the comparison of diagnostic capabilities for determining the prescription of mechanical brain injury using algorithms of statistical (central statistical moments of the 1st - 4th orders), fractal (approximating curves to logarithmic dependences of power spectra) and multifractal (WTMM) analysis of MMI linear birefringence of fibrillar networks of neurons of nervous tissue. Excellent (~95%) accuracy of differential diagnosis of the prescription of mechanical injury has been achieved.
Бесплатно
Статья научная
Mental conditions such as fatigue, distraction, and cognitive overload are known to contribute significantly to traffic accidents. Accurate recognition of these cognitive and emotional states is therefore important for the development of intelligent monitoring systems. In this study, a multimodal emotion recognition framework using electroencephalography (EEG) signals and facial expression features is proposed, with potential applications in driver monitoring. The approach integrates Long Short-Term Memory (LSTM) networks and Transformer architectures for EEG-based temporal feature extraction, along with Vision Transformers (ViT) for facial feature representation. Feature-level fusion is employed to combine physiological and visual modalities, enabling improved emotion classification performance compared to unimodal approaches. The model is evaluated using accuracy, precision, recall, and F1-score metrics, achieving an overall accuracy of 96.38%, demonstrating the effectiveness of multimodal learning. Although the experiments are conducted on general-purpose emotion datasets, the results indicate that the proposed framework can serve as a reliable foundation for driver monitoring applications, such as fatigue, distraction, and cognitive state assessment, in intelligent transportation systems.
Бесплатно
Статья научная
In recent times People commonly display substantial intra-class variability in both appearance and position, making pedestrian recognition difficult. Current computer vision techniques like object identification as well as object classification has given deep learning (DL) models a lot of attention and this application is based on supervised learning, which necessitates labels. Multimodal imaging enables examining more than one molecule at a time, so that cellular events may be examined simultaneously or the progression of these events can be followed in real-time. Purpose of this study is to propose and construct a hybrid machine learning (ML) pedestrian identification model based on multimodal datasets. For pedestrian detection, the input is gathered as multimodal pictures, which are then processed for noise reduction, smoothing, and normalization. Then, the improved picture was categorized using metaheuristic salp cross-modal swarm optimization and optimized using naive spatio kernelized extreme convolutional transfer learning. We thoroughly evaluated the proposed approach on three benchmark datasets for multimodal pedestrian identification that are made accessible to the general public. For several multimodal image-based pedestrian datasets, experimental analysis is done in terms of average precision, log-average miss rate, accuracy, F1 score, and equal error rate. The findings of the studies show that our method is capable of performing cutting-edge detection on open datasets. proposed technique attained average precision of 95%, log-average miss rate of 81%, accuracy of 61%, F1 score of 51%, equal error rate of 59%.
Бесплатно
Multiple Objects Tracking Using CAMShift Algorithm and Implementation of Trip Wire
Статья научная
In this paper we represent Security application which is developed using concepts of Video Analytics. User can draw Trip wire on video stream with help of Mouse Callback events. Using this application user can restrict any area of total video scene. Direction selection for tripping is also a choice of a user. If any undesired moving object cross this drawn trip wire then motion of this moving object is getting detected and also tracked. If object crosses trip wire in the same direction as that of user selected then Alarm Indication will appear on that moving object. OpenCV library functions are used for motion detection and motion tracking. CAMShift algorithm is implemented for tracking. An experimental result shows Motion detection, Motion Tracking and drawn trip wire on video.
Бесплатно
Myanmar Continuous Speech Recognition System Using Convolutional Neural Network
Статья научная
Translating the human speech signal into the text words is also known as Automatic Speech Recognition System (ASR) that is still many challenges in the processes of continuous speech recognition. Recognition System for Continuous speech develops with the four processes: segmentation, extraction the feature, classification and then recognition. Nowadays, because of the various changes of weather condition, the weather news becomes very important part for everybody. Mostly, the deaf people can’t hear weather news when the weather news is broadcast by using radio and television channel but the deaf people also need to know about that news report. This system designed to classify and recognize the weather news words as the Myanmar texts on the sounds of Myanmar weather news reporting. In this system, two types of input features are used based on Mel Frequency Cepstral Coefficient (MFCC) feature extraction method such MFCC features and MFCC features images. Then these two types of features are trained to build the acoustic model and are classified these features using the Convolutional Neural Network (CNN) classifiers. As the experimental result, The Word Error Rate (WER) of this entire system is 18.75% on the MFCC features and 11.2% on the MFCC features images.
Бесплатно
Статья научная
Super resolution is a technique to enhance the scale of image in digital image processing. The single low resolution and multiple low resolution techniques have been used by many researchers in reconstructing high resolution image. The above resolution increasing techniques are researched under spatial and frequency domain. When increased in the resolution of image, it is very important to retain the quality of image, which is the challenging task in the domain of digital image processing. Here in this paper, the super resolution architecture for single low resolution technique has been proposed to reconstruct the high resolution image by combining interpolation and restoration methods in spatial domain. The modified adaptive bilinear interpolation is proposed for interpolation and contra harmonic mean & adaptive median filter are used for restoration of single low resolution image. The experimentation is done on standard data set show that, the results obtained from modified adaptive bilinear interpolation are competitively improved when compare to other existing single low resolution techniques in interpolation domain.
Бесплатно
Neural Network Synchronous Binary Counter Using Hybrid Algorithm Training
Статья научная
Information processing using Neural Network Counter can result in faster and accurate computation of data due to their parallel processing, learning and adaptability to various environments. In this paper, a novel 4-Bit Negative Edge Triggered Binary Synchronous Up/Down Counter using Artificial Neural Networks trained with hybrid algorithms is proposed. The Counter was built solely using logic gates and flip flops, and then they are trained using different evolutionary algorithms, with a multi objective fitness function using the back propagation learning. Thus, the device is less prone to error with a very fast convergence rate. The simulation results of proposed hybrid algorithms are compared in terms of network weights, bit-value, percentage error and variance with respect to theoretical outputs which show that the proposed counter has values close to the theoretical outputs.
Бесплатно
Статья научная
Fractal analysis is currently in full swing in particular in the medical field because of the fractal nature of natural phenomena (vascular system, nervous system, bones, breast tissue ...). For this, many algorithms for estimating the fractal dimension have emerged. Most of them are based on the principle of box counting. In this work we propose a new method for calculating fractal attributes based on contrast homogeneity and energy that have been extracted from gray level co-occurrence matrix. As application we are investigated in the characterization and classification of mammographic images with SuportVectorMachine classifier. We considered in particular images with tumor masses and architectural disorder to compare with normal ones. We calculate, for comparison the fractal dimension obtained by a reference method (triangular prism) and perform a classification similar to the previous. Results obtained with new algorithm are better than reference method (classification rate is 0.91 vs 0.65). Hence new fractal attributes are relevant.
Бесплатно
New Biometric Approaches for Improved Person Identification Using Facial Detection
Статья научная
Biometrics is measurable characteristics specific to an individual. Face detection has diverse applications especially as an identification solution which can meet the crying needs in security areas. While traditionally 2D images of faces have been used, 3D scans that contain both 3D data and registered color are becoming easier to acquire. Before 3D face images can be used to identify an individual, they require some form of initial alignment information, typically based on facial feature locations. We follow this by a discussion of the algorithms performance when constrained to frontal images and an analysis of its performance on a more complex dataset with significant head pose variation using 3D face data for detection provides a promising route to improved performance.
Бесплатно
New Intelligent-based Approach for the Early Detection of Disorders: Use on Rhinological Data
Статья научная
Medical data are characterized by complexity, inaccuracy, heterogeneity, the presence of hidden dependencies, often their distributions are unknown. Correlations between factors of disorders, including clinical data, parameters of time series, patient’s subjective assessments have a high complexity that cannot be fully comprehended by humans anymore. This problem is extremely important especially in case of the early detection of disorders. Machine learning methods are very useful for such detection task. Special area of interest is a problem of breathing disorders. In the paper, author demonstrates the potential use of computational intelligence tools for rhinologic data processing. Implementation of supervised learning techniques will allow improving accuracy of disorders detection as well as decrease medical insurance company expenses. Proposed intelligent-based approach makes it possible to process a variety of heterogeneous data in the medical domain. A combination of conventional and fractal features for time series of rhinomanometric data as well as inclusion of hydrodynamic characteristics of nasal breathing process provides the best accuracy. Such approach may be modified for other breathing disorders detection.
Бесплатно
New Mean-Variance Gamma Method for Automatic Gamma Correction
Статья научная
Gamma correction is an interesting method for improving image quality in uncontrolled illumination conditions case. This paper presents a new technique called Mean-Variance Gamma (MV-Gamma), which is used for estimating automatically the amount of gamma correction, in the absence of any information about environmental light and imaging device. First, we valued every row and column of image pixels matrix as a random variable, where we can calculate a feature vector of means/variances of image rows and columns. After that, we applied a range of inverse gamma values on the input image, and we calculated the feature vector, for each inverse gamma value, to compare it with the target one defined from statistics of good-light images. The inverse gamma value which gave a minimum Euclidean distance between the image feature vector and the target one was selected. Experiments results, on various test images, confirmed the superiority of the proposed method compared with existing tested ones.
Бесплатно
New automatic target recognition approach based on Hough transform and mutual information
Статья научная
This paper presents a new automatic target recognition approach based on Hough transform and mutual information. The Hough transform groups the extracted edge points in edged images to an appropriate set of lines which helps in features extraction and matching processes in both of target and stored database images. This gives an initial indication about realization and recognition between target image and its corresponding database image. Mutual information is used to emphasize the recognition of the target image and its verification with its corresponding database image. The proposed recognition approach passed through five stages which are: edge detection by Sobel edge detector, thinning as a morphological operation, Hough transformation, matching process and finally measuring the mutual information between target and the available database images. The experimental results proved that, the target recognition is realized and gives more accurate and successful recognition rate than other recent recognition techniques which are based on stable edge weighted HOG.
Бесплатно
Next-Gen Market Predictor: Transformed Moving Average Fast-RNN Hybrid with Advanced CNNS
Статья научная
Stock price prediction anticipates future stock prices using historical data and computational models to assist and guide investing decisions. In financial forecasting, accuracy and efficacy in stock price prediction are essential for making better choices. This research describes a hybrid deep learning strategy for improving the extraction and interpretation of the crucial details from stock price time series data. Traditional approaches confront challenges such as computational complexity and nonlinear stock prices. The suggested method pre-processes stock data with Moving Average Z-Transformation, which emphasises long-term trends and reduces fluctuations in the short term. It combines a Transformed Moving Average Fast-RNN Hybrid with Advanced CNNs to create an efficient computational framework. The Enhanced Deep-CNN layer comprises convolutional layers, batch normalisation, leaky ReLU activations, dropout, max pooling and a dense layer. The performance of the model is quantified using metrics including Mean Absolute Error (MAE), Mean Square Error (MSE), Root Mean Square Error (RMSE), and R-squared (R2). It shows superior prediction accuracy with MAEs of 0.28, 0.15, 0.34, 0.17, and 0.13 for Kotak, ICICI, Axis, and SBI, respectively, outperforming previous models. These measurements provide detailed information about the model's predictive skills, proving its ability to improve stock price forecast accuracy significantly.
Бесплатно
Noise Removal From Microarray Images Using Maximum a Posteriori Based Bivariate Estimator
Статья научная
Microarray Image contains information about thousands of genes in an organism and these images are affected by several types of noises. They affect the circular edges of spots and thus degrade the image quality. Hence noise removal is the first step of cDNA microarray image analysis for obtaining gene ex-pression level and identifying the infected cells. The Dual Tree Complex Wavelet Transform (DT-CWT) is preferred for denoising microarray images due to its properties like improved directional selectivity and near shift-invariance. In this paper, bivariate estimators namely Linear Minimum Mean Squared Error (LMMSE) and Maximum A Posteriori (MAP) derived by applying DT-CWT are used for denoising microarray images. Experimental results show that MAP based denoising method outperforms existing denoising techniques for microarray images.
Бесплатно
Noisy Image Decomposition Based On Texture Detecting Function
Статья научная
At present, most of image decomposition models only apply to some ideal images, such as, noise-free, without blurring and super resolution images, and so on. In this paper, they propose a novel decomposition model based on dual method and texture detecting function for noisy image. Firstly, they prove the existence of minimal solutions of the noisy decomposition model functional. Secondly, they write down an alterative implementation algorithm. Finally, they give some numerical experiments, which show that their model can effectively work for Gaussian noisy image decomposition.
Бесплатно
