Visual saliency detection models are widely used in computer vision tasks to mimic the human visual system’s perception of scenes. The part of an image that stands out from its surroundings and captures attention at a glance is referred to as the salient region. This paper presents a comprehensive review of recent advancements in Salient Object Detection (SOD) and its subfields, namely Co-Salient Object Detection (CSD) and RGB-Depth (RGBD) saliency detection. Salient Object Detection refers to techniques that analyze image surroundings to extract prominent regions from the background. Co-saliency detection, on the other hand, focuses on identifying common and salient regions across a group of related images that share similar content. In contrast to traditional saliency detection, RGBD models incorporate both color and depth information to more accurately identify salient objects. All the aforementioned SOD approaches have numerous applications in pattern recognition and computer vision. The concept of saliency detection has garnered significant interest among researchers. However, there remains a need for extensive research to produce highly accurate saliency maps that bridge the gap between perceptual accuracy and computational performance. Although many novel methods have been developed to address these challenges, further efforts are required to enhance the overall accuracy of saliency detection. This review covers a wide range of techniques, from traditional approaches to deep learning-based models. In addition to analyzing various proposed algorithms, the paper provides a comprehensive overview of evaluation metrics used to assess the performance of SOD algorithms. It also explores benchmark datasets commonly used in SOD research and presents both qualitative and quantitative experimental results for SOD and its subfields. Finally, this study examines open research problems, current challenges, and future directions in salient object detection, offering valuable insights and guidance to support future research advancements in the field.

In the era of machine learning we are solving the classification problems by training the labeled classes. But sometimes due to insufficient data in some of the training classes, the system training is inadequate for these minority classes. In this case the output for the classes obtained from the less amount of trained data are miserably inappropriate and biased towards the classes having more data. This problem is known as a class imbalance problem. In such cases, standard classifiers tend to be overpowered by the expansive classes and disregard the little ones. As a result, the performance of machine learning and the deep learning algorithms are also reducing and sometimes highly unacceptable too, mainly if it is related to crucial data like medical and health related. Though various researchers provided some methods to solve this problem but mostly they are problem specific and suitable with the specific classifier only. To find a generalized and effective solution to this problem, we have applied various smote variants for solving the imbalanced factors in dataset and finally improved the performance of the various machine learning and deep learning algorithms. We have experimented and analyzed the effects of SMOTE variants on various machine learning techniques over six standard medical datasets. We have found that SMOTE variants are very effective, and they improve the standard performance measures (Accuracy, Precision, Recall and F1-Score). Additionally, based on our research, it is feasible to determine which smote variation works best with machine learning methods and datasets

For quick criminal activity alert, it became quite obvious to use multiple capture points with cameras, and with multiple capture points, there is need for automated criminal activity alerting systems so the human observer can manage all the CCTV feed in real-time, as its humanly not possible go through that many video feed without a crime slipping out undetected. A deep learning based approach for this task on various network have been researched previously. The researchers have tuned with larger and largest network possible to perform weapon detection for surveillance, though they have achieved more than 90% of accuracy for the task but have to pull largest and complex networks possible. Large deep learning models are costly in both computation and memory for a CCTV device to perform AI workload.There was a gap in studying hybrid approach where deep learning along with machine learning based approach are evaluated for the task. To close this gap, our study employs a hybrid approach that combines machine learning and deep learning methods.Training on a customised dataset was attempted initially. But when implementation proved challenging, the study transitioned to implementing the use of the “OD-weapon detection dataset” that had been collected from GitHub. Different levels of accuracy were achieved on first validation by using deep learning models such as VGG16, VGG19, InceptionNet, and MobileNet, which were maximised by applying this diversified collection of weapon images. Techniques for clustering, fine-tuning, and PCA dimension reduction were used to improve the classification performance

The widespread usage of mobile devices and social media has led to a growing interest in face recognition technology. This study introduces a novel deep ensemble method designed to enhance facial recognition accuracy on a mobile selfie dataset by integrating three pre-trained models, viz. Inception-v3, ResNet-50, and EfficientNet B7 for automatic feature extraction and representation. The approach utilizes feature-level fusion through concatenation, followed by dimensionality reduction via principal component analysis (PCA). Feature optimization is carried out using the Firefly algorithm, and classification is achieved through a soft voting ensemble of classifiers, including Support Vector Machine (SVM), Random Forest, and a Deep Neural Network (DNN). When evaluated on the LFW, UTK face, and Wild Selfie datasets, the proposed method achieved recognition accuracies of 99.76%, 98.92%, and 98.73%, respectively, demonstrating competitive and significantly improved performance over existing models. The results indicate that the system performs effectively in real-world conditions, especially in environments with varying conditions.

Salient Object Detection experiences significant difficulties when trying to identify objects from single haze images due to the deterioration of visibility and low contrast. To subdue this challenge, this study introduces a computational model of visual saliency as a solution. Object detection in hazy environments presents a major challenge due to reduced visibility and contrast. The proposed methodology begins by determining whether an image is hazy, and if so, leverages the Dark Channel Prior (DCP) to extract essential haze-related information. The DCP calculation serves as the basis for subsequent dehazing, achieved through the Multiscale Retinex algorithm. In the dehazing phase, the Multiscale Retinex algorithm is applied to improve image clarity and obtain a dehazed version. This haze-free image is given as input to a trained U-Net architecture, which gives a saliency map that identifies notable and prominent regions within the image. Simultaneously, it undergoes region-based segmentation. The geodesic saliency map is calculated using geodesic distance, considering both spatial proximity and feature similarity. In the final step, the saliency maps generated from the U-Net and geodesic saliency computation are fused to generate the ultimate saliency map. The effectiveness of the suggested method in detecting salient objects in hazy images is supported by the experimental findings, which showcase state-of-the-art performance in dehazing. The integration of DCP, multiscale Retinex, and dual saliency maps enhances both dehazing and object detection, making this method valuable in a variety of computer vision applications, including autonomous driving, video surveillance, and image restoration. The experimental results of AUC and MAE provide confirmation for the effectiveness and accuracy of the saliency computational model that has been proposed.

Salient Object Detection (SOD) can mimic the human vision system by using algorithms that simulate the way how the eye detects and processes visual information. It focuses mainly on the visually distinctive parts of an image, similar to how the human brain processes visual information. The approach proposed in this study is an ensemble approach that incorporates classification algorithm, foreground connectivity and prior calculations. It involves a series of preprocessing, feature generation, selection, training, and prediction using random forest to identify and extract salient objects in an image as a first step. Next, an object proposals map is created for the foreground object. Subsequently, a fusion map is generated using boundary, global, and local contrast priors. In the feature generation step, different edge filters are implemented as the saliency score at edges will be high; additionally, with the use of Gabor’s filter the texture-based features are calculated. The Boruta feature selection algorithm is then used to identify the most appropriate and discriminative features, which helps to reduce the computational time required for feature selection. Ultimately, the initial map obtained from the random forest, along with the fusion saliency maps based on foreground connectivity and prior calculations, is merged to produce a saliency map. This map is then refined using post-processing techniques to acquire the final saliency map. The approach we propose surpasses the performance of 17 cutting-edge techniques across three benchmark datasets, showcasing superior results in terms of precision, recall, and f-measure. The proposed method performs well even on the DUT-OMRON dataset, known for its multiple salient objects and complex backgrounds, achieving a Mean Absolute Error (MAE) value of 0.113. The method also demonstrates high recall values (0.862, 0.923, 0.849 for ECSSD, MSRA-B and DUT-OMRON datasets, respectively) across all datasets, further establishing its suitability for salient object detection. © 2024 by the authors.

Saliency Detection is a crucial undertaking in the realm of vision computing, with a goal to identify the visual prominent regions within an input image. The method of automated saliency identification has caught the interest of various application fields during the last decade. An innovative method is suggested for saliency detection through Conditional Generative Adversarial Networks (CGANs) with a pre-trained U-Net model as the generator. The generated saliency maps are evaluated by the discriminator for authenticity and give feedback to enhance the generator’s ability to generate high-resolution saliency maps. By iteratively training the discriminator and generator networks, the model achieves improved results in finding the salient object. By combining the strengths of conditional generative adversarial networks and the U-Net architecture, our goal is to improve the accuracy and enhance the quality. Once the U-Net model is trained and its weights are saved, we then integrate it into the CGAN framework for salient object detection. The U-Net will serve as part of the generator for the CGAN, responsible for generating saliency maps for input images. The components of CGAN, are trained using adversarial learning to enhance the quality and realism of the resulting saliency maps. Precision, recall, MAE, and F? score measurements are used to evaluate performance. Thorough experiments have been conducted on three challenging saliency detection datasets, our model has demonstrated remarkable performance surpassing the latest models for saliency. Further, faster convergence is observed in our model due to the initialization of the CGAN’s generator using pre-trained U-Net model weights. © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2024.

Recent research shows that Conditional Generative Adversarial Networks (cGANs) are effective for Salient Object Detection (SOD), a challenging computer vision task that mimics the way human vision focuses on important parts of an image. However, implementing cGANs for this task has presented several complexities, including instability during training with skip connections, weak generators, and difficulty in capturing context information for challenging images. These challenges are particularly evident when dealing with input images containing small salient objects against complex backgrounds, underscoring the need for careful design and tuning of cGANs to ensure accurate segmentation and detection of salient objects. To address these issues, we propose an innovative method for SOD using a cGAN framework. Our method utilizes encoder-decoder framework as the generator component for cGAN, enhancing the feature extraction process and facilitating accurate segmentation of the salient objects. We incorporate Wasserstein-1 distance within the cGAN training process to improve the accuracy of finding the salient objects and stabilize the training process. Additionally, our enhanced model efficiently captures intricate saliency cues by leveraging the spatial attention gate with global average pooling and regularization. The introduction of global average pooling layers in the encoder and decoder paths enhances the network's global perception and fine-grained detail capture, while the channel attention mechanism, facilitated by dense layers, dynamically modulates feature maps to amplify saliency cues. The generated saliency maps are evaluated by the discriminator for authenticity and gives feedback to enhance the generator's ability to generate high-resolution saliency maps. By iteratively training the discriminator and generator networks, the model achieves improved results in finding the salient object. We trained and validated our model using large-scale benchmark datasets commonly used for salient object detection, namely DUTS, ECSSD, and DUT-OMRON. Our approach was evaluated using standard performance metrics on these datasets. Precision, recall, MAE and F? score metrics are used to evaluate performance. Our method achieved the lowest MAE values: 0.0292 on the ECSSD dataset, 0.033 on the DUTS-TE dataset, and 0.0439 on the challenging and complex DUT-OMRON dataset, compared to other state-of-the-art methods. Our proposed method demonstrates significant improvements in salient object detection, highlighting its potential benefits for real-life applications. Copyright © 2024 Dhara and Kumar.

In computer vision, image dehazing is a low-level task that employs algorithms to analyze and remove haze from images, resulting in haze-free visuals. The aim of Salient Object Detection (SOD) is to locate the most visually prominent areas in images. However, most SOD techniques applied to visible images struggle in complex scenarios characterized by similarities between the foreground and background, cluttered backgrounds, adverse weather conditions, and low lighting. Identifying objects in hazy images is challenging due to the degradation of visibility caused by atmospheric conditions, leading to diminished visibility and reduced contrast. This paper introduces an innovative approach called Dehaze-SOD, a unique integrated model that addresses two vital tasks: dehazing and salient object detection. The key novelty of Dehaze-SOD lies in its dual functionality, seamlessly integrating dehazing and salient object identification into a unified framework. This is achieved using a conditional Generative Adversarial Network (cGAN) comprising two distinct subnetworks: one for image dehazing and another for salient object detection. The first module, designed with residual blocks, Dark Channel Prior (DCP), total variation, and the multiscale Retinex algorithm, processes the input hazy images. The second module employs an enhanced EfficientNet architecture with added attention mechanisms and pixel-wise refinement to further improve the dehazing process. The outputs from these subnetworks are combined to produce dehazed images, which are then fed into our proposed encoder–decoder framework for salient object detection. The cGAN is trained with two modules working together: the generator aims to produce haze-free images, whereas the discriminator distinguishes between the generated haze-free images and real haze-free images. Dehaze-SOD demonstrates superior performance compared to state-of-the-art dehazing methods in terms of color fidelity, visibility enhancement, and haze removal. The proposed method effectively produces high-quality, haze-free images from various hazy inputs and accurately detects salient objects within them. This makes Dehaze-SOD a promising tool for improving salient object detection in challenging hazy conditions. The effectiveness of our approach has been validated using benchmark evaluation metrics such as mean absolute error (MAE), peak signal-to-noise ratio (PSNR), and structural similarity index measure (SSIM).

Pneumonia ranks among the world's major causes of mortality and is the greatest cause of death for young children. It is an infectious condition that can be fatal, affects one or both lungs and is brought on by harmful bacteria. An accurate and timely diagnosis is essential for managing and treating patients effectively. Radiotherapists with specialized training are needed to assess chest X-rays to diagnose pneumonia. Therefore, creating an automated approach to identify pneumonia would be advantageous to treat the illness, especially in isolated locations quickly. This project offers a novel method for improving chest X-ray image quality, which is then used in conjunction with machine learning approaches to increase the detection accuracy of pneumonia. Subtle details in X-rays can be seen much better using picture-enhancing techniques including sharpening, contrast stretching, and histogram equalization. A VGG net and a convolutional neural network (CNN) model that can accurately diagnose pneumonia is trained using this augmented image dataset. By bridging the gap between conventional X-ray imaging and sophisticated machine learning, the initiative offers a viable approach to the early and accurate detection of pneumonia. Early disease identification is greatly aided by medical imaging, and chest X-rays are a frequent method of identifying lung disorders like pneumonia. This project offers a novel method for improving chest X-ray image quality, which is then used in conjunction with machine learning approaches to increase the detection accuracy of pneumonia. Subtle details in X-rays can be seen much better using picture-enhancing techniques including sharpening, contrast stretching, and histogram equalization. A Convolutional Neural Network (CNN) model that can accurately diagnose pneumonia is trained using this augmented image dataset. By bridging the gap between conventional X-ray imaging and sophisticated machine learning, the initiative offers a viable approach to the early and accurate detection of pneumonia.

The risk of developing COVID-19 and its variants may be higher in those with pre-existing health conditions such as thyroid disease, Hepatitis C Virus (HCV), breast tissue disease, chronic dermatitis, and other severe infections. Early and precise identification of these disorders is critical. A huge number of patients in nations like India require early and rapid testing as a preventative measure. The problem of imbalance arises from the skewed nature of data in which the instances from majority class are classified correct, while the minority class is unfortunately misclassified by many classifiers. When it comes to human life, this kind of misclassification is unacceptable. To solve the misclassification issue and improve accuracy in such datasets, we applied a variety of data balancing techniques to several machine learning algorithms. The outcomes are encouraging, with a considerable increase in accuracy. As an outcome of these proper diagnoses, we can make plans and take the required actions to stop patients from acquiring serious health issues or viral infections.

The Voice User Interface (VUI) for human-computer interaction has received wide acceptance, due to which the systems for speech recognition in regional languages are now being developed, taking into account all of the dialects. Because of the limited availability of the speech corpus (SC) of regional languages for doing research, designing a speech recognition system is challenging. This contribution provides a Parallel Big Bang-Big Crunch (PB3C)-based mechanism to automatically evolve the optimal architecture of LSTM (Long Short-Term Memory). To decide the optimal architecture, we evolved a number of neurons and hidden layers of LSTM model. We validated the proposed approach on Marathi speech recognition system. In this research work, the performance comparisons of the proposed method are done with BBBC based LSTM and manually configured LSTM. The results indicate that the proposed approach is better than two other approaches.

-

-

-

Credit scoring plays a vital role for financial institutions to estimate the risk associated with a credit applicant applied for credit product. It is estimated based on applicants' credentials and directly affects to viability of issuing institutions. However, there may be a large number of irrelevant features in the credit scoring dataset. Due to irrelevant features, the credit scoring models may lead to poorer classification performances and higher complexity. So, by removing redundant and irrelevant features may overcome the problem with large number of features. In this work, we emphasized on the role of feature selection to enhance the predictive performance of credit scoring model. Towards to feature selection, Binary BAT optimization technique is utilized with a novel fitness function. Further, proposed approach aggregated with 'Radial Basis Function Neural Network (RBFN)', 'Support Vector Machine (SVM)' and 'Random Forest (RF)' for classification. Proposed approach is validated on four bench-marked credit scoring datasets obtained from UCI repository. Further, the comprehensive investigational results analysis are directed to show the comparative performance of the classification tasks with features selected by various approaches and other state-of-the-art approaches for credit scoring.

-