Multimedia Tools and Applications最新文献

Various speech processing approaches (e.g., acoustic feature extraction techniques) and Machine Learning (ML) algorithms have been applied to diagnosing Parkinson's disease (PD). However, the majority of these researches have used conventional techniques which obtain a low accuracy rate in diagnosing PD and still need further improvement. Particle Swarm Optimization-Extreme Learning Machine (PSO-ELM), one of the most recent and effective ML techniques, could be considered an accurate strategy in the classification process but has not been applied to solve the problem of PD diagnosis. Thus, in order to enhance the precision of the PD diagnosing, this study employs the PSO-ELM classifier and examines how well it performs on seven feature extraction techniques (basic features, WT (Wavelet Transform), MFCC (Mel Frequency Cepstral Coefficients), bandwidth + formant, intensity parameters, TQWT (Tunable Q-factor Wavelet Transform), and vocal fold features). The PSO-ELM approach has the capability to a) prevents overfitting, b) solve the binary and multi class classification issues, and c) perform like a kernel-based support vector machine with a structure of neural network. Therefore, if the combination of PSO-ELM classifier and appropriate feature extraction technique can improve learning performance, this combination can produce an effective method for identifying PD. In this study, the PD's voice samples have been taken from the Parkinson’s Disease Classification Benchmark Dataset. To discover a useful feature extraction technique to couple with the PSO-ELM classifier, we applied PSO-ELM to each extracted feature with the utilisation of unbalanced and balanced dataset. According to the experimental results, the MFCC features assist the PSO-ELM classifier to attaining its greatest accuracy, up to 97.35% using unbalanced dataset and 100.00% using balanced dataset. This shows that combining PSO-ELM with MFCC can improve learning performance, ultimately creating an effective method for identifying PD.

In the field of ophthalmology, digital images play an important role for automatic detection of various kind of eye diseases. Digital images in the field image enhancement are the first stage to assisting ophthalmologist for diagnosis. As a result, various algorithms, and methods for the enhancement of retinal images have been developed, which may face obstacles that are common in augmentation processes, such as false edges and weak illuminated that obscure image particulars. To eliminate such issues, this paper projected a novel framework for unexposed retinal image. The proposed paper uses multiscale Gaussian function for estimation of illumination layer from unexposed color retinal image and then it is corrected by gamma method. Further to this, the principal component analysis (PCA) is utilized here to generate fused enhance result for unexposed retinal images. Then, contrast limited technique is employed here for further edge and contextual details improvement. When compared to several enhancement-based state-of-the-art procedures, experimental results show that the suggested method produces results with good contrast and brightness. The significance of the proposed method that this method may help ophthalmologists screen for unexposed retinal illnesses more efficiently and build better automated image analysis for healthcare diagnosis.

COVID-19 emerged as a pandemic in December 2019. This virus targets the pulmonary systems of humans. Therefore, chest radiographic imaging is required to monitor effect of the virus, prevent the spread and decrease the mortality rate. Imaging-based testing leads to a high burden on the radiologist manually screening the images. To make the imaging-based method an efficient diagnosis tool, screening automation with minimum human interference is a necessity. It opens numerous challenges for scientists and researchers to develop automatic diagnostic tools for COVID-19 detection. In this paper, we present two speed-enhanced convolutional neural networks (SECNNs) to automatically detect COVID-19 among the X-rays of COVID-19, pneumonia and healthy subjects. For 2-class classification (2CC) and 3-class classification (3CC), we named the models SECNN-2CC and SECNN-3CC respectively. The scope of this work is to highlight the significance and potential of CNN models built from scratch in COVID-19 identification. We conduct six experiments using six different balanced and imbalanced kinds of datasets. In the datasets, All X-rays are from different patients therefore it was more challenging for us to design the models which extract abstract features from a highly variable dataset. Experimental results show that the proposed models exhibit exemplary performance. The highest accuracy for 2CC (COVID-19 vs Pneumonia) is 99.92%. For 3CC (COVID-19 vs Normal vs Pneumonia), the highest accuracy achieved is 99.51%. We believe that this study will be of great importance in diagnosing COVID-19 and also provide a deeper analysis to discriminate among pneumonia, COVID-19 patients and healthy subjects using X-rays.

To evaluate multiple choice question tests, optical forms are commonly used for large-scale exams and these forms are read by the OMR (Optical Mark Recognition) scanners. However, OMR scanners often misinterpret marks that have not been fully erased, which can lead to incorrect readings. To overcome that shortcoming and reduce the time and labor lost in the assessment process, we developed a novel system based on the density of each individual’s markings, providing a more personalized and accurate approach. Instead of reading according to a specific optical form template, a dynamic and flexible structure was generated where users can create own templates and obtain the model that reads according to that template. We also optimized certain aspects of the system for efficiency, such as image memory transfer and QR code reading. These optimizations significantly increase the performance of the OMR scanners. One of the key issues addressed is inaccurate reading of OMR scanners when a student doesn’t fully erase their markings or when markings are faint. After the scanning process, the proposed approach uses a K-means clustering algorithm to classify different density markings. This technique identifies each student’s personal marking density, enabling a more accurate interpretation of their responses. According to the experimental results, we performed 97.7% improvement compared to the misread optics scanned by the conventional OMR devices. In tests performed on 265.816 optical forms, we obtained an accuracy rate of 99.98% and a reading time of 0.12 seconds per optical form.

A film is considered an experience good, as its quality is only revealed after consumption. This situation creates information asymmetry before consumption, prompting producers, who are aware of their film’s quality, to search for methods to signal this. Economic literature specifies that a signal to disclose a product’s quality must be strong, meaning only producers of good-quality films can effectively utilize such a signal. However, a poster represents the most economical signal, and all producers, regardless of film quality, have access to this option. To study whether a poster can signal film quality, we first apply a low-dimensional representation of poster images and cluster them to identify quality-related patterns. We then perform a supervised classification of films into economically successful and unsuccessful categories using a deep neural network. This is based on the hypothesis that higher quality films tend to sell more tickets and that all producers invest in the highest quality poster services. The results demonstrate that a film’s quality can indeed be predicted from its poster, reinforcing its effectiveness as a strong signal. Despite the proliferation of advanced visual media technologies, a simple yet innovative poster remains an effective and appealing tool for signaling film information. Notably, posters can classify a film’s economic success comparably to trailers but with significantly lower processing costs.

Automated detection of cardiovascular diseases based on heartbeats is a difficult and demanding task in signal processing because the routine analysis of the patient’s cardiac arrhythmia is crucial to reducing the mortality rate. Detecting and preventing these deaths requires long-term monitoring and manual examination of electrocardiogram (ECG) signals, which takes a lot of time. This article uses an optimized Vision Transformer technique to effectively detect heart disease. The four key processes are pre-processing input data, feature extraction from pre-processed data, and optimal feature selection and classification to detect heart disease. In the pre-processing phase, single-channel adaptive blind source separation is used for artifact removal and empirical mode decomposition for noise reduction of the ECG signal. After pre-processing, the ECG signal is fed into the Enhanced Pan-Tompkins algorithm (EPTA) and the Hybrid Gabor-Walsh-Hadamard transform (HGWHT) for feature extraction. The extracted feature is selected using a Manifold Grasshopper Optimization algorithm (MGOA). Finally, an Optimized Vision Transformer (OVT) detects heart disease. The experiment is carried out on PTB diagnostic ECG and PTB-XL database, a publicly accessible research datasets. The experiment obtained the following values: accuracy 99.9%, sensitivity 98%, F1 score 99.9%, specificity 90%, processing time 13.254 s, AUC 99.9% and MCC 91% using PTB diagnostic ECG. On the other hand, the proposed method has obtained an accuracy of 99.57%, f1-score of 99.17% and AUC of 99% using PTB-XL dataset. Thus, the overall findings prove that the proposed method outperforms the existing methodology.

To decrease domain shift in cross-domain person re-identification, existing methods generate pseudo labels for training models, however, the inherent distribution between source domain data and the hard quantization loss is ignored. Therefore, a cross-domain person re-identification method based on distribution consistency and multi-label collaborative learning is proposed. Firstly, a soft binary cross-entropy loss function is constructed to constrain the inter-sample relationship of cross-domain transformation, which can ensure the consistency of appearance features and sample distribution, and achieving feature cross-domain alignment. On this basis, in order to suppress the noise of hard pseudo labels, a multi-label collaborative learning network is constructed. The soft pseudo labels are generated by using the collaborative foreground features and global features to guide the network training, making the model adapt to the target domain. The experimental results show that the proposed method has better performance than that of recent representative methods.

Recommender systems play an essential role in decision-making in the information age by reducing information overload via retrieving the most relevant information in various applications. They also present great opportunities and challenges for business, government, education, and other fields. The cold start problem is a significant issue in these systems. If recommender systems fail to provide satisfactory personalized recommendations for new users, the user’s trust can easily be lost. Hence, in this paper, using co-clustering and utilizing user demographic information and the behavioral history of users, a solution to this critical issue for recommending movies is introduced. In the proposed method, in addition to dealing with the problem of relative cold start, the problem of absolute cold start is also addressed. The proposed method was evaluated via two RMSE and MAE criteria, which accordingly has achieved 0.85 and 0.49 on the Movielens dataset and 1.05 and 0.6 on the EachMovie dataset, respectively, according to the number of comments that Cold Start users have registered. Moreover, it achieved 0.9 and 0.55 on the Movielens dataset and 1.42 and 0.89 on the EachMovie dataset, respectively, according to the number of registered comments for the cold start items.

In an acoustic pervasive wireless sensor network (PWSN), the BASE station plays a vital role in gathering and integrating acoustic sensor data from various nodes, including end and router devices tracking time-driven events. The semantic BASE station is crucial in the IoT landscape as it consolidates data from these networks, enabling thorough sentiment analysis of acoustic signals and yielding insights across domains. A semantic processor at the BASE station is essential for an energy-efficient and intelligent PWSN, managing data collection, integration, signal feature extraction, and publication for model training and sentiment analysis. This paper introduces a novel approach to designing a semantic BASE station, focusing on ontology generation, evaluation, and updates to bolster pervasive wireless sensors in capturing and depicting events and time through an ontological framework. The study addresses challenges in efficiently gathering, integrating, and processing acoustic data from pervasive nodes, proposing a semantic processor at the BASE station to enhance feature extraction and metadata publication. The semantic organization of feature-extracted labeled metadata enables the analysis of comprehensive machine learning (ML) applications such as sentiment analysis, type detection, and environment detection by generating confusion matrix. Evaluation includes performance metrics (NEEN, LSNS, BDAS) as well as accuracy, precision, sensitivity, and specificity for sentimental data analysis to validate the proposed technique’s efficacy.

People transmit millions of digital images daily over various networks, where securing these images is a big challenge. Image encryption is a successful approach widely used in securing digital images while transmitting. Researchers developed different encryption techniques that focus on securing individual images. Recently, encryption of multiple images has gained more interest as an emerging encryption approach. In this paper, we proposed a four-tier technique for multiple image encryption (MIE) to increase the transmission speed and improve digital image security. First, we attached the plain images to create an augmented image. Second, the randomized augmented image is obtained by randomly changing the position of each plain image. Third, we scrambled the randomized augmented image using the zigzag pattern, rotation, and random permutation between blocks. Finally, we diffuse the scrambled augmented image using an Altered Sine-logistic-based Tent map (ASLT). We draw a flowchart, write a pseudo-code, and present an illustrative example to simplify the proposed method and make it easy to understand. Many experiments were performed to evaluate this Four-Tier technique, and the results show that this technique is extremely effective and secure to withstand various attacks.