International Journal of Pattern Recognition and Artificial Intelligence

Glaucoma is an eye disease that causes loss of vision and blindness by damaging a nerve in the back of the eye called optic nerve. The optic nerve collects the visual information from the eyes and transmits to the brain. Glaucoma is mainly caused by an abnormal high pressure in the eyes. Over time, the increased pressure can erode the tissues of optic nerve, leading to vision loss or blindness. If it is diagnosed in advance, then only it can prevent the vision loss. To diagnose the glaucoma, it must accurately differentiate between the optic disc (OD), optic cup (OC), and the retinal nerve fiber layer (RNFL). The segmentation of the OD, OC, and RNFL remains a challenging issue under a minimum contrast image of boundaries. Therefore, in this study, an innovative method of Hybrid Symbiotic Differential Evolution Moth-Flame Optimization (SDMFO)-Multi-Boost Ensemble and Support Vector Machine (MBSVM)-based segmentation and classification framework is proposed for accurately detecting the glaucoma disease. By using Group Search Optimizer (GSO), the affected parts of the OD, OC and RNFL are segmented. The proposed SDMFO-MBSVM method is executed in MATLAB site, its performance is analyzed with three existing methods. From the comparison, the accuracy of the proposed method in OD segmentation gives better results of 3.37%, 4.54% and 2.22%, OC segmentation gives better results of 2.22%, 3.37% and 4.54%, and RNFL segmentation gives the better results of 3.37%, 97.21% and 5.74%.

Site road alignment is one of the main elements of architectural environmental design [B. Holdsworth, Refocus 6(1) (2005) 58–60]. It is difficult for traditional methods to achieve both qualitative exploration and quantitative optimization objectives, and optimization algorithms can only optimize quantitative objectives. On the other hand, the shape of the site road is subject to various other human interventions besides the designer, which is a multi-objective problem. Based on the idea of meta-design, this paper proposes a new method for “form-finding” of site roads. This method develops the architecture and components of an expert system, which can handle both qualitative and quantitative objectives in conceptual design, and can realize various human interventions. Component development combines site specification, recursive algorithm, and genetic algorithm, which treats human intervention as a factor of variation, allows for qualitative exploration of different preferences, and finally demonstrates the capabilities of this new approach through case studies.

In the field of motor imagery (MI) recognition, poor generalization and low recognition performance are major challenges. An MI recognition method based on semi-supervised learning and multi-source transfer learning is proposed. In this approach, samples are transferred from some source domains to the target domain using the multi-source transfer learning method. The source domains selection method based on distribution similarity is designed to select source domains with similar distribution to the target domain, and samples with high information entropy are selected from these source domains for transfer. In this regard, we propose a semi-supervised learning labeling method for labeling the unlabeled samples of the target domain, which utilizes the labeling information from a few labeled samples without increasing the labeling cost. The sample confidence measurement method and the dynamic adjustment mechanism are proposed to ensure labeling accuracy and minimize the influence of mislabeled samples. A fusion classification model can identify the new sample in the target domain. As a measure of the effectiveness of the proposed method, four types of MI from the BCI Competition IV dataset 2A were used to evaluate the recognition ability, and the outcomes confirmed an excellent recognition performance as well as a superior training efficiency when compared with the currently used methods.

Speech-driven user interfaces are becoming more common in our lives. To interact with such systems naturally and effectively, machines need to recognize the emotional states of users and respond to them accordingly. At the heart of the emotion recognition research done to this end lies the emotion representation that enables machines to learn and predict emotions. Speech emotion recognition studies use a wide range of low-to-high-level acoustic features for representation purposes such as LLDs, their functionals, and BoAW. In this paper, we present a new method for extracting a novel set of high-level features for classifying emotions. For this purpose, we (1) reduce the dimension of discrete-time speech signals, (2) perform a quantization operation on the new signals and assign a distinct symbol to each quantization level, (3) use the symbol sequences representing the signals to extract discriminative patterns that are capable of distinguishing different emotions from each other, and (4) generate a separate set of features for each emotion from the extracted patterns. Experimental results show that pattern features outperform Energy, Voicing, MFCC, Spectral, and RASTA feature sets. We also demonstrate that combining the pattern-based features and the acoustic features further improves the classification performance.

With the explosively growing amount of online information, recommender system becomes an important tool to help users efficiently find their desired information. In this paper, we propose a Graph Neural Network Social Recommendation Based on Coupled Influence by analyzing the social influence of 2-level friends (CI-GNNSR). First, we mine the user’s historical rating information and second-degree social information. Then, to learn the feature representation of users and movies, multiple Graph Attention Networks (GAT) are used to model the user-movie Graph and social network Graph. Our algorithm uses an attention-based memory network to learn the interest influence representation between users and their collaborative friends, which can distinguish the related factors among different users’ friends. The experiment results show that CI-GNNSR enhances the accuracy of recommendation by considering users’ social influence factors from multiple perspectives.

Traditional machine learning projects have revolved around training the model with the help of previously observed data to be able to predict output for future unknown data. In the current scenario, when the data generated are huge, centralized training of the model becomes inefficient. Hence, distributed approach with client server model has to be used for training the models. This introduces data handling and critical data privacy issues. This paper concentrates on Federated learning (FL) which builds a model for the server by aggregating the parameters obtained from the local models of the client devices. The research work focuses on design and evaluation of three new FL algorithms against the average of the performances of the local models. The evolved approach considering weights of the local models proportional to accuracy of the local model is found to be the most accurate and better than the centralized approach. The evaluation is done using three different algorithms belonging to regression and classification on multiple datasets. It is observed that there is only one round of communication between the clients and server required in the federated learning setup to achieve the benchmarked accuracy set by the centralized setup. This is a considerable development and state-of-the-art approach to reduce communication and computation costs.

The traditional computer simulation landscape painting has the defect that it cannot completely present the painting, the application research of local color simulation method of landscape painting based on deep learning adversarial networks is proposed. Based on the improved generative adversarial networks design, the semantic label hierarchical classification design of landscape paintings is carried out. Through the training and design of generative adversarial networks, the generator and discriminator of landscape painting are designed, and the semantic segmentation algorithm of landscape painting of generative adversarial networks is proposed. Finally, a simulation experiment test is carried out on the local color simulation of landscape painting. The results show that this application method is better than the traditional computer simulation method, it can fully reflect the realism of landscape art painting and the integrity of the picture itself. The texture detail clarity of the generated map is stronger than that of the traditional one, and the semantic content is more accurate. It has important practical reference value.

One of the most common approaches through which people communicate is facial expressions. A large number of features documented in the literature were created by hand, with the goal of overcoming specific challenges such as occlusions, scale, and illumination variations. These classic methods are then applied to a dataset of facial images or frames in order to train a classifier. The majority of these studies perform admirably on datasets of images shot in a controlled environment, but they struggle with more difficult datasets (FER-2013) that have higher image variation and partial faces. The nonuniform features of the human face as well as changes in lighting, shadows, facial posture, and direction are the key obstacles. Techniques of deep learning have been studied as a set of methodologies for gaining scalability and robustness on new forms of data. In this paper, we look at how well-known deep learning techniques (e.g. GoogLeNet, AlexNet) perform when it comes to facial expression identification, and propose an enhanced hybrid deep learning model based on STN for facial emotion recognition, which gives the best feature extraction and classification in one go and maximizes the accuracy for a large number of samples on FERG, JAFFE, FER-2013, and CK+ datasets. It is capable of focusing on the main parts of the face and attaining extensive development over preceding fashions on the FERG, JAFFE, CK+ datasets, and the more challenging one namely FER-2013.

Icons and screenshots are important media displayed in game distribution platforms for providing a brief understanding of the game content to the customers. In this study, we develop ensemble convolutional neural networks for icon and screenshot analysis as three applications: an automatic genre classification, a similar game searching, and a recognition quality assessment. First, the genre classifier is developed using 154358 images from 18 030 games in 17 genres. The proposed genre classifiers achieve 40.5% and 47.6% accuracies for classifying a single icon and a single screenshot, which outperform the average performance of the human testers. The accuracy can be boosted to 54.2% by aggregating results from every image of the game. The Grad-CAM is applied to analyze what models learned. Then, the feature extraction part trained by this task is transferred to the other two applications. For the similar game searching, a dissimilarity of two images is directly computed by the Euclidean distance in the feature space. We define a dissimilarity between two games which are sets of multiple images based on their image-pairwise dissimilarity. The results show that the features are successfully transferred, and the model seems to be able to cluster the games with a similar gameplay and differentiate them from the other gameplays even if they come from the same genre. For the third application, we develop a system for quality assessment of game images based on the correctness of viewers’ understanding of game content by combining multiple models from three different problem definitions. Our system can identify good-genre-representing game images which most of the human testers can recognize their genre correctly with 75.0% accuracy for icons and 76.2% accuracy for screenshots.

Upsampling operation is necessary for semantic segmentation and other pixel-level prediction tasks. Among the commonly used upsampling operations, some are too simple to effectively recover the spatial details lost during downsampling process, and some are too complex and have high computation complexity. In real-world applications, it is critical to achieve high accuracy and maintain real-time inference speed. Therefore, an efficient upsampling operation is essential for these tasks. In this paper, we introduce efficient adaptive upsampling module (EAUM) for real-time semantic segmentation. Inspired by dynamic filter networks, EAUM adaptively predicts the kernel weight of each point in the upsampled feature map according to the corresponding points in the input feature map. To reduce computational cost, EAUM decomposes the spatial information and channel information required for upsampling. The proposed EAUM shows impressive performance on Cityscapes and CamVid benchmarks. Specifically, DenseENet with EAUM outperforms the baseline by 1.4% $(73.6\%\to 75.0\%)$ and 1.6% $(75.2\%\to 76.8\%)$ in accuracy with a slight drop in inference speed on Cityscapes test dataset.

The internet of things (IoT) is a rapidly expanding network of smart digital devices that can communicate with one another and be controlled remotely over the internet. Moreover, IoT devices are cheap and can be used to control and monitor activities remotely. Due to this reason, IoT is widely used in the applications of a smart city. Moreover, the smart devices that are used in IoT-based smart city applications are used to gather information from devices, humans, and other sources for analyzing purposes. Hence, it is crucial to conduct the face recognition process to ensure the safety of the city. Several works were conducted by the researchers to recognize the face accurately. Typically, the effectiveness of achieving face recognition is still an intricate one. To tackle those issues, we have proposed a novel condition convolutional neural network (condition-CNN)-based bee foraging learning (BFL)-based particle swarm optimization (PSO) algorithm (CCNNBFLPSO). To recognize the facial images from the face image datasets, the proposed CCNNBFLPSO model is used. To ensure the prediction accuracy condition, CNN uses the conditional probability weight matrix (CPWM) to estimate the higher and lower class level of image features. Meanwhile, the learning of CPWM can be performed by utilizing the adopted BPL-PSO approach. For experimental purposes, we have taken three datasets namely the CVL face database, the MUCT database, and the CMU multi-PIE face database. The training time and the training accuracy are analyzed for all the three datasets, and comparative studies are performed with state-of-art works such as LBPH, FoL TDL, and TPS approaches. The training and validation loss functions are analyzed with baseline CNNs, B-CNN, and condition-CNN. The proposed approach accomplishes better face recognition accuracy and F1-score of about 99.9% and 99.9%, respectively.

In this paper, a system based on image descriptor and Local Histogram Concatenation (LHC) for finger vein recognition is introduced. The LHC of image descriptors such as LBP, LDP CLBP cannot be inverted back to the original images, therefore they can provide good security if stored as enrolled data. On the other hand, the technique of LHC does not depict spatial information, therefore it is expected to be less sensitive to image misalignment if a measure such as the histogram difference $(d{X}^2)$ is used for recognition. The use of histogram difference makes the system more robust to misalignment compared to the pixel-by-pixel-based measures such as the Hamming Distance (HD). The approach of LHC is implemented by dividing the image descriptor into non-overlapped grids, then the histogram within each grid is calculated and concatenated with the histograms of the preceding grids and finally, the concatenated histograms of each two images are compared using $d{X}^2$ measure. Two datasets, UTFVP and SDUMLA-HMT, are used for testing the performance of the system. The results have shown that the Identification Recognition Rate (IRR) is improved when LHCs of the image descriptors with $(d{X}^2)$ measure are used compared to the use of only the image descriptors with HD measure. For UTFVP dataset, the IRR values were 97.44%, 95% and 98.37% when LHC and $(d{X}^2)$ were used with LBP, LDP and CLBP, respectively, while these values were 89.44%, 92.63% and 92.92% when only LBP, LDP and CLBP with HD were used. For SDUMLA-HMT dataset, the IRR values of the system were 98.43%, 98.69% and 98.85% when LHC and $(d{X}^2)$ were used with LBP, LDP and CLBP, respectively, while these values were 97.6%, 98.24% and 97.27% when only the image descriptors LBP, LDP and CLBP with HD were used.

Vision-based road friction coefficient estimation received extensive attention in the field of road maintenance and autonomous driving. However, the current mainstream coarse-grained friction estimation methods are basically based on image classification tasks. This makes it difficult to deal with complex road conditions in changing weathers. Many models can correctly predict the friction coefficients of the road as a whole in consistent and simple road conditions, but perform poorly otherwise. The existing image benchmarks in this field rarely consider the above problems as well, which limits the comparable evaluations of different models. Therefore, in this paper, we first construct a challenging pixel-level friction coefficient estimation dataset WRF-P to evaluate model performances under mixed road conditions. Then, we propose a friction coefficient estimation method based on weakly supervised learning and uncertainty estimation to realize pixel-level road friction prediction with low annotation cost. The model outperforms existing weakly supervised methods and reaches 39.63% mIOU on the WRF-P dataset. The WRF-P dataset will be made publicly available at https://github.com/blackholeLFL/The-WRF-dataset soon.

Because of the various earth return pathways in practical projects, calculating the wide-band earth return impedance of submarine cable is problematic. Using the dynamic Hertz magnetic potential, the formula for estimating the earth return impedance has been obtained based on the three-layer structure of air, sea water, and seafloor. Furthermore, in the frequency range of 50 $\sim$ 10 MHz, the enhanced complex image approach and numerical integration method have been developed to compute the earth return impedance. The suggested iterative complex image method outperforms the classic complex image method in terms of accuracy and computation speed, with a global fitting error of less than 10⁻⁶. Finally, the impact of various models on the real and imaginary parts of submarine cable wide-band earth return impedance has been investigated. The results demonstrate that the actual component has a more complicated variation, whereas the imaginary part is clearly affected by the skin effect, with a bigger difference in the high-frequency range. The method presented in this research can be utilized as a supplement to existing methods, as well as a reference methodology for analyzing the influence of over-voltage on submarine cable engineering.