Temporal information has useful features for recognizing facial expressions. However, to manually design requires a lot of effort. In this paper, reduce effort, deep learning technique, which is regarded as tool automatically extract from raw data, adopted. Our network based on two different models. The first extracts temporal appearance image sequences, while the other geometry landmark points. These models are combined using new integration method in order boost performance expression...

Face recognition under viewpoint and illumination changes is a difficult problem, so many researchers have tried to solve this problem by producing the pose- illumination- invariant feature. Zhu et al. [26] changed all arbitrary pose images frontal view image use for In scheme, preserving identity while rotating crucial issue. This paper proposes new deep architecture based on novel type of multitask learning, which can achieve superior performance in target-pose face from an identity. The...

Unsupervised domain adaptation (UDA) methods for learning invariant representations have achieved remarkable progress. However, most of the studies were based on direct from source to target and suffered large discrepancies. In this paper, we propose a UDA method that effectively handles such We introduce fixed ratio-based mixup augment multiple intermediate domains between domain. From augmented-domains, train source-dominant model target-dominant complementary characteristics. Using our...

A catastrophic forgetting problem makes deep neural networks forget the previously learned information, when learning data collected in new environments, such as by different sensors or light conditions. This paper presents a method for alleviating problem. Unlike previous research, our does not use any information from source domain. Surprisingly, is very effective to less of domain, and we show effectiveness using several experiments. Furthermore, observed that occurs between mini-batches...

In this paper, we present ResNet-based vehicle classification and localization methods using real traffic surveillance recordings. We utilize a MIOvision dataset, which comprises 11 categories including variety of vehicles, such as bicycle, bus, car, motorcycle, so on. To improve the performance, exploit technique called joint fine-tuning (JF). addition, propose dropping CNN (DropCNN) method to create synergy effect with JF. For localization, implement basic concepts state-of-the-art region...

In this paper, we propose an efficient lane detection algorithm for departure detection; is suitable low computing power systems like automobile black boxes. First, extract candidate points, which are support to a hypotheses as two lines. step, Haar-like features used, and enables us use integral image remove computational redundancy. Second, our verifies the hypothesis using defined rules. These rules based on assumption that camera installed at center of vehicle. Finally, if detected, then...

Expanding the domain that deep neural network has already learned without accessing old data is a challenging task because networks forget previously information when learning new from domain. In this paper, we propose less-forgetful method for expansion scenario. While existing adaptation techniques solely focused on adapting to domains, proposed technique focuses working well with both and domains needing know whether input or First, present two naive approaches which will be problematic,...

In remote sensing, numerous unlabeled images are continuously accumulated over time, and it is difficult to annotate all the data. Therefore, a self-supervised learning technique that can improve recognition rate using data will be useful for sensing. This letter presents contrastive with smoothed representation sensing based on SimCLR framework. well-known characteristic within short distance might semantically similar usually used. Our algorithm this knowledge, simultaneously utilizes...

In recent decades, multi-objective evolutionary algorithms (MOEAs) have been evaluated on artificial test problems with unrealistic characteristics, leading to uncertain conclusions about their efficacy in real-world applications. To address this issue, a few benchmark suites comprising proposed for MOEAs, encompassing numerous and select many-objective problems. Given the distinct challenges posed by optimization (MaOPs) inherent difficulty, it is crucial develop suite that includes many...

Deep learning is considered to be a breakthrough in the field of computer vision, since most world records recognition tasks are being broken. In this paper, we try apply such deep techniques recognizing facial expressions that represent human emotions. The procedure our expression system as follows: First, face detected from input image using Haar-like features. Second, network used for faces. step, two different networks can neural and convolutional network. Consequently, compared...

Federated learning (FL) allows UAVs to collaboratively train a globally shared machine model while locally preserving their private data. Recently, the FL in edge-aided unmanned aerial vehicle (UAV) networks has drawn an upsurge of research interest due bursting increase heterogeneous data acquired by and need build global with privacy; however, critical issue is how deal non-independent identically distributed (non-i.i.d.) nature ensuring convergence learning. To effectively address this...

Temporal information can provide useful features for recognizing facial expressions. However, to manually design requires a lot of effort. In this paper, reduce effort, deep learning technique which is regarded as tool automatically extract from raw data, adopted. Our network based on two different models. The first extracts temporal geometry landmark points, while the other appearance image sequences . These models are combined in order boost performance expression recognition. Through...

In this paper, we propose a novel rigid motion segmentation algorithm called randomized voting (RV). This is based on epipolar geometry, and computes score using the distance between feature point corresponding line. accumulated utilized for final grouping. Our basically deals with two frames, so it also applicable to two-view problem. For evaluation of our algorithm, Hopkins 155 dataset, which representative test set segmentation, adopted, consists three motions. has provided most accurate...

Fermented vegetable juices have gained attention due to their various beneficial effects on human health. In this study, we employed gas chromatography–mass spectrometry, direct infusion-mass and liquid spectrometry identify useful metabolites, lipids, carotenoids in juice (VJ) fermented with Lactobacillus plantarum HY7712, HY7715, helveticus HY7801, Bifidobacterium animalis ssp. lactis HY8002. A total of 41 24 4 were detected the non-fermented VJ (control). The lycopene, α-carotene,...

Studies comparing the detection of clean mucosal areas in capsule endoscopy (CE) using human judgment versus artificial intelligence (AI) are rare. This study statistically analyzed gastroenterologist judgments and AI results. Three hundred CE video clips (100 patients) were prepared. Five gastroenterologists classified into 3 groups (≥75% [high], 50%–75% [middle], < 50% [low]) according to their subjective cleanliness. Visualization scores calculated an algorithm based on predicted...

Object detection is a crucial research topic in the fields of computer vision and artificial intelligence, involving identification classification objects within images. Recent advancements deep learning technologies, such as YOLO (You Only Look Once), Faster-R-CNN, SSDs (Single Shot Detectors), have demonstrated high performance object detection. This study utilizes YOLOv8 model for real-time environments requiring fast inference speeds, specifically CCTV automotive dashcam scenarios....

In this article, we propose a novel model for facial micro-expression (FME) recognition. The proposed basically comprises transformer, which is recently used computer vision and has never been FME A transformer requires huge amount of data compared to convolution neural network. Then, use motion features, such as optical flow late fusion complement the lack dataset. method was verified evaluated using SMIC CASME II datasets. Our approach achieved state-of-the-art (SOTA) performance 0.7447...

Over the last few years, deep learning has produced breakthrough results in many application fields including speech recognition, image understanding and so on. We try to techniques for real-time facial expression recognition instead of hand-crafted feature-based methods. The proposed system can recognize human emotions based on expressions using a webcam. It detect faces users with distance 2~3m TV environment. And it determine whether user is feeling happiness, sadness, surprise, anger,...