• 대한전자공학회:학술대회논문집
• /
• 대한전자공학회 2000년도 ITC-CSCC -2
• /
• pp.823-826
• /
• 2000

Motion detection is an important step in computer vision and image processing. Traditional motion detection systems are classified into two categories, namely, feature based and gradient based. In feature based motion detection, features in consecutive frames are detected and matched. Gradient based methods assume that the intensity varies linearly and locally. The method, which we propose, is neither feature nor gradient based but uses the electric field theory. The pixels in an image are modeled as point charges and motion is detected by using the variations between the two electric fields produced by the charges corresponding to the two images.

• 한국멀티미디어학회논문지
• /
• 제17권10호
• /
• pp.1141-1149
• /
• 2014

Recently, various methods to effectively eliminate the noise are researched in image processing techniques. However, the conventional noise filtering techniques, which remove most of the noise, are less efficient for remained noise detection after filtering due to exploiting no face feature information. In this paper, we proposed a feature vector generation technique in the mouth region by distinguishing and revising the remained noise through gradient correction, when the outline is extracted after performing noise filtering.

• 전기전자학회논문지
• /
• 제19권4호
• /
• pp.526-534
• /
• 2015

In this paper, we studied the use of the foreground and background features and SVM classifier to improve the accuracy of offline handwritten numeral recognition. The foreground features are two directional features: directional gradient feature by Kirsch operators and directional stroke feature by local shrinking and expanding operations, and the background feature is concavity feature which is extracted from the convex hull of the numeral, where the concavity feature functions as complement to the directional features. During classification of the numeral, these three features are combined to obtain good discrimination power. The efficiency of our scheme is tested by recognition experiments on the handwritten numeral database CENPARMI, where SVM classifier with RBF kernel is used. The experimental results show the usefulness of our scheme and recognition rate of 99.10% is achieved.

• Advances in Energy Research
• /
• 제6권2호
• /
• pp.91-101
• /
• 2019

This paper proposes an efficient data-driven approach to build models for predicting energy consumption in buildings. Data used in this research is collected by installing humidity and temperature sensors at different locations in a building. In addition to this, weather data from nearby weather station is also included in the dataset to study the impact of weather conditions on energy consumption. One of the main emphasize of this research is to make feature selection independent of domain knowledge. Therefore, to extract useful features from data, two different approaches are tested: one is feature selection through principal component analysis and second is relative importance-based feature selection in original domain. The regression model used in this research is gradient boosting regression and its optimal parameters are chosen through a two staged coarse-fine search approach. In order to evaluate the performance of model, different performance evaluation metrics like r2-score and root mean squared error are used. Results have shown that best performance is achieved, when relative importance-based feature selection is used with gradient boosting regressor. Results of proposed technique has also outperformed the results of support vector machines and neural network-based approaches tested on the same dataset.

• 전자공학회논문지
• /
• 제51권11호
• /
• pp.66-72
• /
• 2014

Histogram of oriented gradient (HOG) 특징은 영상 기반 보행자 인식에서 널리 사용되고 있다. HOG 특징을 이용한 보행자 인식의 인식률을 높이는데 가장 중요한 역할을 하는 것은 보간 기술이다. HOG 특징 연산에 보간 기술을 적용하기 위해서는 각 픽셀의 기울기 방향에 가장 근접한 두 개의 기울기 방향 bin과 가중치를 계산해야 한다. 따라서 본 논문에서는 HOG 특징 연산에 적용하기 위한 효율적인 기울기 방향 bin 및 가중치 연산 회로를 제안한다. 제안하는 회로는 탄젠트 함수와 나눗셈 연산을 피하기 위해 미리 계산된 값을 테이블로 지정하여 사용하였으며, 탄젠트 함수와 가중치 값의 특성을 이용함으로써 회로 내 테이블의 크기를 최소화하였다. 또한 처리 속도 향상을 위해 파이프라인 구조를 적용하였으며, 효율적인 coarse 및 fine 탐색 방법을 적용하여 각 픽셀에 대한 기울기 방향 bin과 가중치를 두 클락 사이클 내에 계산한다. 본 논문에서 제안하는 회로는 $1^{\circ}$ 단위로 기울기 방향을 계산하여 기울기 방향 bin과 가중치를 모두 결정하기 때문에 HOG 특징을 위한 보간 기술에 적용되어 높은 인식률을 제공하기 위해 사용될 수 있다.

• 정보학연구
• /
• 제12권4호
• /
• pp.25-32
• /
• 2009

Iris recognition is a biometric technology which can identify a person using the iris pattern. It is important for the iris recognition system to extract the feature which is invariant to changes in iris patterns. Those changes can be occurred by the influence of lights, changes in the size of the pupil, and head tilting. This paper is appropriate for the embedded environment using local gradient histogram embedded system using iris feature extraction methods have implement. The proposed method enables high-speed feature extraction and feature comparison because it requires no additional processing to obtain the rotation invariance, and shows comparable performance to the well-known previous methods.

• Journal of Information Processing Systems
• /
• 제17권2호
• /
• pp.385-398
• /
• 2021

The complexity of deep learning models affects the real-time performance of gesture recognition, thereby limiting the application of gesture recognition algorithms in actual scenarios. Hence, a residual learning neural network based on a deep convolutional neural network is proposed. First, small convolution kernels are used to extract the local details of gesture images. Subsequently, a shallow residual structure is built to share weights, thereby avoiding gradient disappearance or gradient explosion as the network layer deepens; consequently, the difficulty of model optimisation is simplified. Additional convolutional neural networks are used to accelerate the refinement of deep abstract features based on the spatial importance of the gesture feature distribution. Finally, a fully connected cascade softmax classifier is used to complete the gesture recognition. Compared with the dense connection multiplexing feature information network, the proposed algorithm is optimised in feature multiplexing to avoid performance fluctuations caused by feature redundancy. Experimental results from the ISOGD gesture dataset and Gesture dataset prove that the proposed algorithm affords a fast convergence speed and high accuracy.

• 한국정보전자통신기술학회논문지
• /
• 제10권2호
• /
• pp.141-146
• /
• 2017

본 논문에서는 저전력 영상 특징 추출 하드웨어 설계를 위한 하드웨어 폴딩 기법 기반 저면적 Gradient magnitude 연산기 구조를 제안한다. 하드웨어 복잡도를 줄이기 위해 Gradient magnitude 벡터의 특징을 분석하여 기존 알고리즘을 하드웨어를 공유하여 사용할 수 있는 알고리즘으로 변경하여 Folding 구조가 적용될 수 있도록 했다. 제안된 하드웨어 구조는 기존 알고리즘의 특징을 최대한 이용했기 때문에 데이터 품질의 열화가 거의 없이 구현될 수 있다. 제안된 하드웨어 구조는 Altera Quartus II v16.0 환경에서 Altera Cyclone VI (EP4CE115F29C7N) FPGA를 이용하여 구현되었다. 구현 결과, 기존 하드웨어 구조를 이용하여 구현한 연산기와의 비교에서 41%의 logic elements, 62%의 embedded multiplier 절감 효과가 있음을 확인했다.

• 한국멀티미디어학회논문지
• /
• 제7권12호
• /
• pp.1639-1649
• /
• 2004

In this paper, a method for detection of forged signatures based on spectral analysis of directional gradient density function and a weighted fuzzy classifier is proposed. The well defined outline of an incoming signature image is extracted in a preprocessing stage which includes noise reduction, automatic thresholding, image restoration and erosion process. The directional gradient density function derived from extracted signature outline is highly related to the overall shape of signature image, and thus its frequency spectrum is used as a feature set. With this spectral feature set, having a property to be invariant in size, shift, and rotation, a weighted fuzzy classifier is evaluated for the verification of freehand and random forgeries. Experiments show that less than 5% averaged error rate can be achieved on a database of 500 signature samples.

• 제어로봇시스템학회:학술대회논문집
• /
• 제어로봇시스템학회 2004년도 ICCAS
• /
• pp.663-667
• /
• 2004

This paper presents new pruning and learning methods for the fuzzy rule-based classifier. The structure of the proposed classifier is framed from the fuzzy sets in the premise part of the rule and the Bayesian classifier in the consequent part. For the simplicity of the model structure, the unnecessary features for each fuzzy rule are eliminated through the iterative pruning algorithm. The quality of the feature is measured by the proposed correctness method, which is defined as the ratio of the fuzzy values for a set of the feature values on the decision region to one for all feature values. For the improvement of the classification performance, the parameters of the proposed classifier are finely adjusted by using the gradient descent method so that the misclassified feature vectors are correctly re-categorized. The cost function is determined as the squared-error between the classifier output for the correct class and the sum of the maximum output for the rest and a positive scalar. Then, the learning rules are derived from forming the gradient. Finally, the fuzzy rule-based classifier is tested on two data sets and is found to demonstrate an excellent performance.