• Journal of Electrical Engineering and Technology
• /
• v.13 no.6
• /
• pp.2468-2478
• /
• 2018

We propose the Sparse Feature Convolutional Neural Network (SFCNN) to reduce the volume of convolutional neural networks (CNNs). Despite the superior classification performance of CNNs, their enormous network volume requires high computational cost and long processing time, making real-time applications such as online-training difficult. We propose an advanced network that reduces the volume of conventional CNNs by producing a region-based sparse feature map. To produce the sparse feature map, two complementary region-based value extraction methods, cluster max extraction and local value extraction, are proposed. Cluster max is selected as the main function based on experimental results. To evaluate SFCNN, we conduct an experiment with two conventional CNNs. The network trains 59 times faster and tests 81 times faster than the VGG network, with a 1.2% loss of accuracy in multi-class classification using the Caltech101 dataset. In vehicle classification using the GTI Vehicle Image Database, the network trains 88 times faster and tests 94 times faster than the conventional CNNs, with a 0.1% loss of accuracy.

• Journal of the Korea Institute of Information and Communication Engineering
• /
• v.18 no.11
• /
• pp.2727-2732
• /
• 2014

Recently, the researches of object tracking and recognition using Microsoft's Kinect are being actively studied. In this environment human hand detection and tracking is the most basic technique for human computer interaction. This paper proposes a method of improving the accuracy of the detected hand region's boundary in the cluttered background. To do this, we combine the hand detection results using the skin color with the extracted depth image from Kinect. From the experimental results, we show that the proposed method increase the accuracy of the hand region detection than the method of detecting a hand region with a depth image only. If the proposed method is applied to the sign language or gesture recognition system it is expected to contribute much to accuracy improvement.

• Journal of Intelligence and Information Systems
• /
• v.12 no.1
• /
• pp.33-41
• /
• 2006

This paper presents a method that automatically recognizes products into model type, which it flows with the conveyor belt. The specific interlaced image are occurred by moving image when we use the NTSC based camera. It is impossible to process interlaced images, so a suitable post-processing is required. For the purpose of this processing, after it remove interlaced images using de-interlacing method, it leads rectangle region of object by thresholding. And then, after rectangle region is separated into several blocks through edge detection, we calculate pixel numbers per each block, re-classify using its average, and classify products into model type. Through experiments, we know that the proposed method represent high classification ratio.

• Journal of the Semiconductor & Display Technology
• /
• v.16 no.3
• /
• pp.111-115
• /
• 2017

Terahertz waves are superior to conventional X-ray or Magnetic Resonance Tomography(MRI), and the amount of information that can be transmitted is as large as thousands of times that conventional X-ray or MRI. In addition, Terahertz waves have great performance in analyzing an object which have some layered structure. By using this advantage, we can extract the letters of a page by analyzing information such as absorption amount and reflection amount by irradiating a closed book with pulses of various frequencies within gap of a terahertz wave. However, in the image of each page using the Terahertz wave might be obtained various kinds of noise and the different character occlusion region. So, to extract letters from the terahertz image, we must take the noise and occlusion region away. We have been working to enhancement the image quality in various ways, and keep on studying de-noising processing for enhancement about the image quality and high resolution. Finally, we also keep on studying about OCR(Optical Character Recognition) technology, which based on pattern matching technique, to read letters.

• Journal of Korea Multimedia Society
• /
• v.25 no.5
• /
• pp.670-675
• /
• 2022

This study proposes an improved binarization method to improve image recognition rate. The research goal is to minimize the information loss that occurs during the binarization process, and to transform the object of the original image that cannot be determined through the transformation process into an image that can be judged. The proposed method uses a stepwise segmentation method of an image and divides blocks using prime numbers. Also, within one block, a trapezoidal type of fuzzy is applied. The fuzzy trapezoid is binarized by dividing the brightness histogram area into three parts according to the degree of membership. As a result of the experiment, information loss was minimized in general images. In addition, it was found that the converted binarized image expressed the object better than the original image in the special image in which the brightness region was tilted to one side.

• Journal of the Korean Institute of Telematics and Electronics B
• /
• v.33B no.3
• /
• pp.176-185
• /
• 1996

Transformin the huge size of drawings into a suitable format for CAD system and recognizng the contents of drawings are the major concerans in the automated analysis of engineering drawings. This paper proposes some methods for text/graphics separation, symbol extraction, vectorization and symbol recognition with the object of applying them to electronic cirucit drawings. We use MBR (Minimum bounding rectangle) and size of isolated region on the drawings for separating text and graphic regions. Characteristics parameters such as the number of pixels, the length of circular constant and the degree of round shape are used for extracting loop symbols and geometric structures for non-loop symbols. To recognize symbols, nearest netighbor between FD (foruier descriptor) of extractd symbols and these of classification reference symbols is used. Experimental results show that the proposed method can generate compact vector representation of extracted symbols and perform the scale change and rotation of extracted symbol using symbol vectorization. Also we achieve an efficient searching of circuit drawings.

• Journal of Korea Multimedia Society
• /
• v.12 no.6
• /
• pp.804-814
• /
• 2009

In this paper, a novel system for avatar motion controlling by tracking face is presented. The system is composed of three main parts: firstly, LCS (Local Cluster Searching) method based face feature detection algorithm, secondly, HMM based feature points recognition algorithm, and finally, avatar controlling and animation generation algorithm. In LCS method, face region can be divided into many small piece regions in horizontal and vertical direction. Then the method will judge each cross point that if it is an object point, edge point or the background point. The HMM method will distinguish the mouth, eyes, nose etc. from these feature points. Based on the detected facial feature points, the 3D avatar is controlled by two ways: avatar orientation and animation, the avatar orientation controlling information can be acquired by analyzing facial geometric information; avatar animation can be generated from the face feature points smoothly. And finally for evaluating performance of the developed system, we implement the system on Window XP OS, the results show that the system can have an excellent performance.

• International Journal of High-Rise Buildings
• /
• v.9 no.4
• /
• pp.335-342
• /
• 2020

This article comes up with a new method which is based on the visual characteristic of the objects and machine learning technology to achieve semi-automated recognition of the personnel, machine & materials of the construction sites. Balancing the real-time performance and accuracy, using Faster RCNN (Faster Region-based Convolutional Neural Networks) with transfer learning method appears to be a rational choice. After fine-tuning an ImageNet pre-trained Faster RCNN and testing with it, the result shows that the precision ratio (mAP) has so far reached 67.62%, while the recall ratio (AR) has reached 56.23%. In other word, this recognizing method has achieved rational performance. Further inference with the video of the construction of Huoshenshan Hospital also indicates preliminary success.

• Tunnel and Underground Space
• /
• v.33 no.4
• /
• pp.281-298
• /
• 2023

Continuous research efforts are being devoted to unmanned mobile platforms for lunar exploration. There is an ongoing demand for real-time information processing to accurately determine the positioning and mapping of areas of interest on the lunar surface. To apply deep learning processing and analysis techniques to practical rovers, research on software integration and optimization is imperative. In this study, a foundational investigation has been conducted on real-time analysis of virtual lunar base construction site images, aimed at automatically quantifying spatial information of key objects. This study involved transitioning from an existing region-based object recognition algorithm to a boundary box-based algorithm, thus enhancing object recognition accuracy and inference speed. To facilitate extensive data-based object matching training, the Batch Hard Triplet Mining technique was introduced, and research was conducted to optimize both training and inference processes. Furthermore, an improved software system for object recognition and identical object matching was integrated, accompanied by the development of visualization software for the automatic matching of identical objects within input images. Leveraging satellite simulative captured video data for training objects and moving object-captured video data for inference, training and inference for identical object matching were successfully executed. The outcomes of this research suggest the feasibility of implementing 3D spatial information based on continuous-capture video data of mobile platforms and utilizing it for positioning objects within regions of interest. As a result, these findings are expected to contribute to the integration of an automated on-site system for video-based construction monitoring and control of significant target objects within future lunar base construction sites.

• Journal of the Korean Institute of Telematics and Electronics D
• /
• v.34D no.7
• /
• pp.88-96
• /
• 1997

As the need of automatized system has been increased recently together with the development of industrial and military technologies, the adaptive real-time target detection technologies that can be embedded on vehicles, planes, ships, robots and so on, are hgihly demanded. Accordingly, this paper proposes a novel approach to detect and segment the moving targets using the binary phase extraction joint transform correlator (BPEJTC), the advanced image subtraction filter and convex hull processing. The BPEJTC which was used as a target detection unit mainly for target tracking compensating the camera movement. The target region has been detected by processing the successful three frames using the advanced image subtraction filter, and has become more accurate by applying the developed convex hull filter. As shown by some experimental results, it is expected that the proposed approaches for compensation of the camera movement and segmentationof of target region, can be used for th emissile guiddance, aero surveillance, automatic inspectin system as well as the target detection unit of automatic target recognition system that request adaptive real-time processing.