Earticle

We propose Multi-Scale Attention Kolmogorov?Arnold Network (MA-KANet), a novel retinal vessel segmentation framework addressing fine-scale structures and low contrast challenges. Our method integrates Multi-scale Dynamic Fusion (MDF) with 3D convolutional interactions to prevent information loss during downsampling, and Scale-Context Attention Fusion (SCAF) for dynamic feature recalibration in cluttered regions. Kolmogorov?Arnold Network units in decoder stages capture nonlinear dependencies beyond traditional convolutions. MA-KANet achieves state-of-the-art results: 98.80% AUC on DRIVE, 99.05% AUC on CHASEDB1, and 99.17% AUC on FIVES dataset, demonstrating superior performance and generalization across diverse vessel patterns, establishing new benchmarks for retinal vessel segmentation.

Radio maps are essential for electromagnetic spectrum awareness, supporting communication network optimization and spectrum management. However, existing methods often lack sufficient accuracy in complex propagation environments. To address this, we propose RMG-SRGAN, an improved radio map generation network based on an enhanced super-resolution generative adversarial network. The generator incorporates a Multi-scale Attention Aggregation (MAA) module that strengthens feature representation using multi-scale fusion and dual-path attention in spatial and channel dimensions. The discriminator includes a Feature Enhancement (FE) module to boost discriminative power through multi-stage feature processing.In our evaluation, we prioritize physical fidelity and structural reliability over generic perceptual metrics. Consequently, we employ Root Mean Square Error (RMSE) to quantify the precision of predicted signal strength and the F1-Score to assess the classification accuracy of coverage zones versus blind spots. Extensive experiments on the RadioMapSeer dataset demonstrate that RMG-SRGAN achieves state-of-the-art performance, securing the lowest RMSE and highest F1-Score compared to existing baselines.

Biometrics as a tool for information security has been used in various applications. Feature-level fusion is widely used in the design of multi-biometric systems due to its advantages in increasing recognition accuracy and security. However, most existing multi-biometric systems that use feature-level fusion assign each biometric trait an equal proportion when combining features from multiple sources. For example, multi-biometric systems with two biometric traits commonly adopt a 50?50 feature proportion setting, which means that fused feature data contains half elements from each biometric modality. In this paper, we investigate the impact of feature proportion on the matching performance of multi-biometric systems. By using a fingerprint and face based multi-biometric system that applies feature-level fusion, we employ a random projection based transformation and a proportion weight factor. By adjusting this weight factor, we show that allocating unequal proportions to features from different biometric traits yields different matching performance. Our experimental results indicate that optimal performance, achieved with unequal feature proportions, could be better than the performance obtained with the commonly used 50?50 feature proportion. Therefore, the impact of feature proportion, which has been ignored by most existing work, should be taken into account and more study is required as to how to make feature proportion allocation benefit the performance of multi-biometric systems.

Support vector machines (SVMs), a machine learning technique, has been applied to not only binary classification problems such as bankruptcy prediction but also multi-class problems such as corporate credit ratings. However, in general, the performance of SVMs can be easily worse than the best alternative model to SVMs according to the selection of predictors, even though SVMs has the distinguishing feature of successfully classifying and predicting in a lot of dichotomous or multi-class problems. For overcoming the weakness of SVMs, this study has proposed an approach for selecting features for multi-class SVMs that utilize the impurity measures of classification trees. For the selection of the input features, we employed the C4.5 and CART algorithms, including the stepwise method of discriminant analysis, which is a well-known method for selecting features. We have built a multi-class SVMs model for credit rating using the above method and presented experimental results with data regarding S&P 500 companies.

Pothole detection remains a critical challenge in road maintenance and safety management, as potholes deteriorate road surfaces, compromise vehicle safety, and increase maintenance costs. Traditional pothole detection methods relying on manual inspection or simple image processing are often labor-intensive, prone to human error, and lack adaptability to varying road conditions. Meanwhile, modern approaches utilizing single-stage object detectors such as YOLO variants have provided real-time detection capabilities but tend to suffer in accurately localizing potholes at higher Intersection over Union (IoU) thresholds, especially when faced with the irregular shapes and scale variability characteristic of real-world potholes. To overcome these limitations, a multi-stage detection framework based on Cascade Region-based Convolutional Neural Network (Cascade R-CNN) with a ResNet-50 backbone and a Feature Pyramid Network (FPN) was developed. This framework employs progressive bounding box refinement through multiple detection stages with increasingly strict IoU thresholds, resulting in improved localization precision. The model was trained and evaluated on a meticulously curated dataset of more than 30,000 images featuring diverse pothole instances. It achieves a mean Average Precision (mAP) of 0.653 across IoU thresholds from 0.5 to 0.95, surpassing the baseline Faster RCNN by 4.3 points and outperforming YOLOv8 by 5 points. On an NVIDIA RTX 4090 GPU, the proposed model runs at approximately 80–90 frames per second, which enables nearreal- time execution and renders it practical for integration into automated road inspection and maintenance systems. These results indicate that the proposed Cascade R-CNN framework offers a robust and effective solution for high-accuracy pothole detection, addressing the shortcomings of existing detection methods in complex road environments.

본 논문에서는 인간의 시각인지 형태와 유사한 결과를 갖는 Saliency map의 정확성과 신뢰성을 향상시키기 위해 학습한 다중 특징을 기반으로 개선된 saliency map 방법을 제안한다. 기존의 Saliency map 생성에서 색상 기반의 돌출 영역 추정 시 발생하는 역 선택이나 부분손실 등의 부정확한 결과가 나오는 것을 보완하기 위해 제안하는 방법은 학습 기반의 다중 특징 데이터를 생성하였다. 원 영상에서의 색상 패턴과 특이성을 갖는 영역의 구별과정을 거쳐 영상에서 고려될 특성들을 분석하고, LAB 색 공간 기반의 색상 분석을 이용한 유사 돌출 영역 정의와 특이성 영역의 조합으로 학습 데이터를 구성한다. 구성된 학습 데이터와 주파수, 색상, 초점정보 등의 low level feature로 구한 돌출 정보를 결합한 뒤 최종 saliency map을 구하기 위해 재구성 과정을 거쳐 부정확한 saliency 영역을 최소화하도록 하였다. 실험을 위해 Ground truth 이미지를 각 실험 결과와 비교하여 precision-recall 및 F-Measure 값을 구한 결과 기존 알고리즘에 비해 7%, 29%의 향상된 결과를 나타내었다.

In this paper, we propose a method for generating improved saliency map by learning multiple features to improve the accuracy and reliability of saliency map which has similar result to human visual perception type. In order to overcome the inaccurate result of reverse selection or partial loss in color based salient area estimation in existing salience map generation, the proposed method generates multi feature data based on learning. The features to be considered in the image are analyzed through the process of distinguishing the color pattern and the region having the specificity in the original image, and the learning data is composed by the combination of the similar protrusion area definition and the specificity area using the LAB color space based color analysis. After combining the training data with the extrinsic information obtained from low level features such as frequency, color, and focus information, we reconstructed the final saliency map to minimize the inaccurate saliency area. For the experiment, we compared the ground truth image with the experimental results and obtained the precision-recall value.

본 연구는 인공지능 기법 중 다중레이블 속성선택 방법을 적용하여 복잡한 경영환경에서 의사결정의 효과성을 증대시키는 방안을 설명한다. 인공지능 기반의 의사결정 시스템은 의사결정자의 선택과 판단을 돕거나, 대신하는 중요한 역할을 한다. 더욱이 최근 인공지능을 중심으로 한 비즈니스 의사결정은 기업의 성장 동력으로 평가받는데, 이를 위해서는 효과적인 의사결정 방법이 수반되어야 한다. 이에 본 연구는 의미 있는 속성값을 선별하는 CFS-BR(이진연관성 접근 기반의 상관관계 속성선택 모델)을 제안하여, 효과적인 의사결정을 지원하는 것을 돕는다. 예시데이터와 실증데이터의 분석 결과, CFS-BR은 유의미한 속성을 최상우선선별 알고리즘 기반으로 최상의 조합을 선별하므로 효율적 의사결정을 지원할 수 있고, 기존의 다중 레이블 속성선택 방법과 비교하였을 때 정확도가 높은 것으로 보아 효과적인 의사결정을 증대시키는데 유용하다.

Management decision-making based on artificial intelligence(AI) plays an important role in helping decision-makers. Business decision-making centered on AI is evaluated as a driving force for corporate growth. AI-based on accurate analysis techniques could support decision-makers in making high-quality decisions. This study proposes an effective decision-making method with the application of multi-label feature selection. In this regard, We present a CFS-BR (Correlation-based Feature Selection based on Binary Relevance approach) that reduces data sets in high-dimensional space. As a result of analyzing sample data and empirical data, CFS-BR can support efficient decision-making by selecting the best combination of meaningful attributes based on the Best-First algorithm. In addition, compared to the previous multi-label feature selection method, CFS-BR is useful for increasing the effectiveness of decision- making, as its accuracy is higher.

최근 딥러닝 기술의 비약적인 발전으로 인해 극도로 사실적인 합성 미디어인 딥페이크의 제작이 용이해졌다. 특히 GAN(generative adversarial network)과 같은 생성 모델의 발전은 일반인도 쉽게 고품질의 위조 콘텐츠를 생성할 수 있는 환경을 조성하였으며, 이러한 기술의 발전은 개인의 프라이버시 침해, 허위정보 유포, 금융 사기 등 심각한 사회적 문제를 야기하고 있다. 새로운 딥페이크 기술로 생성한 콘텐츠가 확산될 때 가설 기반 콘텐츠의 정보 분포를 활용하여 자가 진화가 가능한 딥페이크 탐지 시스템 기술 개발 높은 정확도를 기반으로 음성과 이미지 그리고 영상 등 다양한 모달리티를 통합하여 탐지할 수 있는 딥페이크 탐지 연구가 필요하며, 본 논문에서는 다중 모달 특성 학습에 기반한 새 로운 생성형 딥페이크 탐지 프레임워크를 제안한다. 제안된 방법은 이미지의 공간적 특성, 비디오의 시간적 특성, 그리 고 오디오 특성을 통합적으로 활용하는 다중 스트림 신경망 구조를 기반으로 하며, 교차 주의집중 메커니즘을 도입하여 각 모달리티 간의 상호 정보를 효과적으로 활용하여, 생성형 딥페이크 탐지를 개선하였다.

The recent rapid development of deep learning technology has made it easier to produce deepfakes, which are extremely realistic synthetic media. In particular, the development of generative models such as GAN (generative adversarial network) has created an environment in which even ordinary people can easily create high-quality counterfeit content, and this development of technology is causing serious social problems such as invasion of personal privacy, dissemination of false information, and financial fraud. In this study, a novel deepfake detection framework is proposed based on multi-modal feature learning. The proposed method is designed with a multi-stream neural network that comprehensively utilizes the spatial attributes of images, the temporal dynamics of videos, and the distinct characteristics of audio. Furthermore, a cross-attention mechanism is introduced to effectively utilize the mutual information among different modalities.

본 논문은 가상공간 제작을 위해 다수의 카메라로 영상을 촬영했을 때, 영상 특성 차이를 감소시키는 방법 을 제안하였다. 각각 8 개의 본체와 렌즈를 교차 장착하여 64 개의 영상을 사용하였다. 영상 분석은 히스토그램과 픽셀 분포 값의 표준 편차를 분석 비교하였다. 분석결과, 동일 기종의 카메라임에도 불구하고, 렌즈 혹은 이미지 센서에 따라 각각 다른 영상 특성을 보여주었다. 본 논문에서는 이러한 차이를 보정하기 위해 영상의 전체 밝기 값의 분포를 조절하였다. 시뮬레이션 결과, 평균 편차가 최대 (실내 : 6.89, 실외 : 24.23) 이었으나, 시뮬레이션 진행 후 편차가 거의(실내 : 최대 0.42, 실외 : 최대 : 2.73) 없는 영상을 얻었다. 추후에는 영상 밝기 분포보다 정밀한 영상 분석 방법을 연구하고 적용할 것이다.

In this paper, we propose a method to reduce the difference in image characteristics when multiple camera images are captured for virtual space production. Sixty-four images were used by cross-mounting eight bodies and lenses, respectively. Image analysis compares and analyzes the standard deviation of the histogram and pixel distribution values. As a result of the analysis, it shows different image characteristics depending on the lens or image sensor, though it is a camera of the same model. In this paper, we have adjusted the distribution of the overall brightness value of the image to compensate for this difference. As a result, the average deviation was the maximum of (Indoor: 6.89, outdoor: 24.23), we obtained images with almost no deviation (Indoor: maximum 0.42, outdoor: maximum: 2.73). In the future, we will study and apply more accurate image analysis methods than image brightness distribution.

본 논문은 다중 에이전트 강화학습 방식을 사용하여 입력 데이터로부터 분류에 효과적인 특징 집합을 찾아내는 방식을 제안한다. 기계 학습 분야에 있어서 분류에 적합한 특징들을 찾아내는 것은 매우 중요하다. 데이터에는 수많은 특징들이 존재할 수 있으며, 여러 특징들 중 일부는 분류나 예측에 효과적일 수 있지만 다른 특징들은 잡음 역할을 함으 로써 올바른 결과를 생성하는 데에 오히려 악영향을 줄 수 있다. 기계 학습 문제에서 분류나 예측 정확도를 높이기 위한 특징 선택은 매우 중요한 문제 중 하나이다. 이러한 문제를 해결하기 위해 강화학습을 통한 특징 선택 방법을 제시한다. 각각의 특징들은 하나의 에이전트를 가지게 되며, 이 에이전트들은 특징을 선택할 것인지 말 것인지에 대한 여부를 결정 한다. 에이전트들에 의해 선택된 특징들과 선택되지 않은 특징들에 대해서 각각 보상을 구한 뒤, 보상에 대한 비교를 통해 에이전트의 Q-value 값을 업데이트 한다. 두 하위 집합에 대한 보상 비교는 에이전트로 하여금 자신의 행동이 옳은지에 대한 판단을 내릴 수 있도록 도와준다. 이러한 과정들을 에피소드 수만큼 반복한 뒤, 최종적으로 특징들을 선별한다. 이 방법을 통해 Wisconsin Breast Cancer, Spambase, Musk, Colon Cancer 데이터 세트에 적용한 결과, 각각 0.0385, 0.0904, 0.1252, 0.2055의 정확도 향상을 보여주었으며, 최종적으로 0.9789, 0.9311, 0.9691, 0.9474 의 분류 정확도를 보여주었다. 이는 우리가 제안한 방법이 분류에 효과적인 특징들을 잘 선별하고 분류에 대한 정확도를 높일 수 있음을 보여준다.

In this paper, we propose a method for finding feature subsets that are effective for classification in an input dataset by using a multi-agent reinforcement learning method. In the field of machine learning, it is crucial to find features suitable for classification. A dataset may have numerous features; while some features may be effective for classification or prediction, others may have little or rather negative effects on results. In machine learning problems, feature selection for increasing classification or prediction accuracy is a critical problem. To solve this problem, we proposed a feature selection method based on reinforced learning. Each feature has one agent, which determines whether the feature is selected . After obtaining corresponding rewards for each feature that is selected, but not by the agents, the Q-value of each agent is updated by comparing the rewards. The reward comparison of the two subsets helps agents determine whether their actions were right. These processes are performed as many times as the number of episodes, and finally, features are selected. As a result of applying this method to the Wisconsin Breast Cancer, Spambase, Musk, and Colon Cancer datasets, accuracy improvements of 0.0385, 0.0904, 0.1252 and 0.2055 were shown, respectively, and finally, classification accuracies of 0.9789, 0.9311, 0.9691 and 0.9474 were achieved, respectively. It was proved that our proposed method could properly select features that were effective for classification and increase classification accuracy.

본 논문에서는 유도 전동기의 결함을 조기에 검출하고 진단하기 위해 이산 코사인 변환과 특이치 분해를 이용한 결함 특징 벡터 추출 방법을 제안하였고, 추출한 특징 벡터를 다층 서포트 벡터 머신의 입력으로 이용하여 유도 전동기의 결함 을 유형별로 분류하는 기법을 기술하였다. 모의실험 결과, 제안한 방법은 5가지의 결함 데이터에 대해 거의 100%의 분류 성능을 보였다.

This paper proposes a method of feature vector extraction using discrete cosine transform(DCT) and singular value decomposition(SVD) for early detection of faults in an induction motor. This, the extracted feature vector is used as an input of multi-layer support vector machine(MLSVM) to classify each fault of the induction motor. Experimental results show that the proposed method achieves 100% accuracy to classify among 5 different faults.

In this paper, we propose a FACET-based feature point extraction method and an image registration method for image fusion based on multi-sensors. The most important part of image registration in image fusion is the extraction of feature points in IR and VIS images. The proposed method of extracting feature points uses the FACET-based filter in an IR image and uses Harris corner detector in a VIS image because it is difficult to find common characteristics and correlations in the IR and VIS images. The proposed method of image registration consists of 5 stages: 1) extraction of the feature points, 2) detection of correspondence points by using mutual information (MI), 3) a 2D projective transformation, 4) interpolation to integers of a pixel coordination, 5) a similar measurement by using normalized mutual information (NMI). The final outcome of the fusion image expresses a pseudo color based on hue (H), saturation (S), and intensity (I). Experimental results showed that the proposed method provides robust and accurate registration results. The expression of a pseudo color image after image fusion showed methods for enhanced target detection.

최근 사물인터넷(IoT) 기술의 고도화로 스마트 홈, 스마트 팩토리, 원격 의료 등 다양한 도메인에서 데이터 수집이 비약적 으로 증가하고 있다. 그러나 이러한 확산과 동시에 DDoS, 데이터 변조, 봇넷 등 사이버 공격 기법이 지능화 및 다변화되면서 보안 위협 또한 심화되고 있다. IoT 환경에서는 이기종 프로토콜과 고차원적인 특징(Feature) 분포를 지니고 있어, 기존의 단 일 탐지 모델로는 연산 복잡성 해소와 실시간 침해 대응에 한계가 있다. 본 논문에서는 이러한 문제를 해결하기 위해 다차원 특징 중요도(Feature Importance) 기반의 조건부 다중 전문가 모델(Conditional Multi-Expert Model)을 제안한다. 제안된 모델 은 입력 데이터의 특징별 중요도를 정량적으로 분석하고 최적의 추론 경로를 동적으로 설정함으로써 탐지 성능을 높이고 연 산자원을 최적화한다. 실험 결과, 제안 모델은 0.9995의 정확도, 0.9948의 F1-Score를 달성하여 기존 단일 모델(Random Forest) 대비 우수한 탐지 성능을 입증하였으며, 특히 재현율(Recall) 0.94%p 향상, 추론시간 약 2.6배 향상시킴으로써 자원 제 약이 심한 IoT 환경에서 실시간 대응이 가능한 수준의 탐지성능과 보안 신뢰성을 확인하였다.

With the recent advancement of Internet of Things (IoT) technology, data collection is increasing exponentially across various domains, including smart homes, smart factories, and telemedicine. However, alongside this expansion, security threats are intensifying as cyber-attack techniques such as DDoS, data manipulation, and botnets become more sophisticated and diversified. Given that IoT environments encompass heterogeneous protocols and high-dimensional feature distributions, conventional single detection models face limitations in reducing computational complexity and achieving real-time incident response. To address these challenges, this paper proposes a Conditional Multi-Expert Model based on high-dimensional feature importance. The proposed model quantitatively analyzes the importance of each feature in the input data and dynamically establishes the optimal inference path, thereby enhancing detection performance and optimizing computational resources. Experimental results demonstrate that the proposed model achieves an accuracy of 0.9995 and an F1-score of 0.9948, proving superior detection performance compared to the conventional single model (Random Forest). In particular, by improving the recall by 0.94%p and accelerating the inference time by approximately 2.6 times, this study confirms a level of detection capability and security reliability capable of real-time response in resource-constrained IoT environments.

본 연구는 형태학적 특징 정량화와 딥러닝 기반 다중 시점 분석을 결합한 송이버섯 품질 등급 자동 분류 시스템을 제안한다. 송이버섯은 형태적 특징에 따라 등급이 결정되나, 전통적 육안 검사는 주관성과 일관성 문제가 있으며, 기존 단일 시점 딥러닝 방법은 3차원 정보 손실로 인한 분류 한계를 보였다. 본 연구는 이중 카메라 시스템을 활용하여 466쌍의 정면-측면 이미지 데이터셋을 구축하고, 규칙 기반의 YOLOv8n 형태 분석 모듈과 딥러닝 기반의 Dual-ConvNeXt 분류 모듈, Dual-EfficientNet 분류 모듈의 3가지 모듈을 Majority Voting 방식으로 앙상블하여 5등급 분류를 수행하였다. 제안 시스템은 94.51%의 정확도를 달성하여 단일 시점 최고 성능(87.91%) 대비 6.6%p, 형태 지표 기반 방식(74.73%) 대비 19.78%p 향상되었다. 본 연구는 다중 시점 정보 융합의 효과를 입증하였으며, 송이버섯 자동 선별의 실용화와 다른 농산물 품질 평가로의 확장 가능성을 제시한다.

This study proposes an automated matsutake mushroom quality grading system that combines rule-based morphological feature quantification with multi-view deep learning analysis using a dual-camera setup. While matsutake mushrooms are graded based on morphological characteristics, traditional visual inspection suffers from subjectivity and inconsistency, and existing single-view deep learning methods encounter classification limitations due to three-dimensional information loss. A dataset of 466 frontal-lateral image pairs was constructed, and five-grade classification was performed by integrating three modules — a rule-based YOLOv8n morphological analysis module, a Dual-ConvNeXt classification module, and a Dual-EfficientNet classification module — via Majority Voting ensemble. The proposed system achieves 94.51% accuracy, improving 6.6 percentage points over the best single-view method (87.91%) and 19.78 percentage points over the morphological approach (74.73%). This research demonstrates the effectiveness of multi-view information fusion and presents the potential for practical implementation of automated matsutake sorting and extension to quality assessment of other agricultural products.

Because of the feature points can describe the local characteristics of the image in a reasonable manner, effective use of feature point of content based image retrieval become the current hot issues in the field of computer vision. Aiming at this problem, we put forward a kind of combination clustering based on feature points, a new method of image retrieval. The method includes the combination of feature point clustering algorithm and based on the algorithm of local color histogram construction strategy. With the existing and local color histogram retrieval method based on feature points, compared to the method can effectively solve the current method of feature point location information and feature point center relying too much on the problem. Subjectivity and as a result of the manual annotation image accuracy, the traditional image retrieval methods cannot meet the needs of the user. Multidimensional indexing technology is only from the perspective of how to improve the indexing algorithm to adapt to the large-scale database to consider a problem, in content-based image retrieval. Our research combines the advantages of the semantic analysis and kernel clustering which will enhance the performance of the traditional image retrieval methods and strengthen the feasibility of the algorithm.

In view of the greater changes of posture, illumination, expression and scene in reality environment have a strong impact on wild face recognition algorithm to identify performance problem, and puts forward a kind of linear discriminant analysis side information (SILD) algorithm on hyperplane fusion of learning prototype. First of all, using support vector machine (SVM) to weak tag of data-concentrated sample is expressed as the middle-level characteristics of prototype hyperplane, using a learning combination coefficient to select sparse support vector set from untagged conventional data set; then, under the constraints of the combination sparse coefficient of SVM model, by using Fisher linear discriminant criterion to maximize discriminant ability of untagged data set, and using the iterative optimization algorithm to solve the objective function; in the end, using SILD for feature extraction, cosine similarity measure to complete the final face recognition. In two general face data sets of wild face recognition (LFW) and YouTube, it makes comparison of PHL+SILD method and low-level features + SILD method on some characteristics, such as strength, LBP, Gabor feature and Block Gabor feature, average accuracy, area under the curve (AUC) and entire error rate (EER). The validity and reliability of the proposed algorithm is verified by the experiments.

Image classification is an important task in computer vision. The methods based on spatial information generally employ some low-level features for image classification, such as gray scale, color, texture and location. It is difficult for vision system to understand and the single feature is too limited to obtain correct classification results. In this paper, an algorithm based on multi-kernel feature learning is proposed and used for image classification. First, the kernel function is used to produce a kernel descriptor, which aggregates the pixel attributes into patch-level features; Then, through the multi-kernel learning, these descriptors are further aggregated to obtain hierarchical multi-feature descriptors; Finally, the label of each image is given by the fusion strategy of on multi-classifiers, which effectively utilizes the advantages of multi-kernel learning and takes the complementary among the classifiers into account. The experimental results show that the proposed method is efficient in promoting the classification results.

Classification of facial expressions is a great challenge in the area of computer vision. Using single feature models, the percentage of recognition is considerably low even in controlled conditions of capturing. To improve the accuracy of the facial emotion recognition and classification, a new method is proposed in this paper based on the fusion of features extracted from different techniques. Local features are extracted using Directional Local Binary Patterns (DLBP) and global features are extracted using Discrete Cosine Transform (DCT) from facial expression images. Principal Components Analysis (PCA) is used to reduce the dimensions of extracted features. Weighted summation and PCA fusion methods are used to fuse the local and global features extracted from facial images. RBF neural network is used as classifier for classification of facial images into six emotions (surprise, fear, sad, joy, anger and disgust). Cohn-Kanade database is used to evaluate the proposed method. The results of proposed algorithm yield better recognition rate of 97% in comparison with existing methods.