• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.13 no.7
• /
• pp.3599-3619
• /
• 2019

In human activity recognition system both static and motion information play crucial role for efficient and competitive results. Most of the existing methods are insufficient to extract video features and unable to investigate the level of contribution of both (Static and Motion) components. Our work highlights this problem and proposes Static-Motion fused features descriptor (SMFD), which intelligently leverages both static and motion features in the form of descriptor. First, static features are learned by two-stream 3D convolutional neural network. Second, trajectories are extracted by tracking key points and only those trajectories have been selected which are located in central region of the original video frame in order to to reduce irrelevant background trajectories as well computational complexity. Then, shape and motion descriptors are obtained along with key points by using SIFT flow. Next, cholesky transformation is introduced to fuse static and motion feature vectors to guarantee the equal contribution of all descriptors. Finally, Long Short-Term Memory (LSTM) network is utilized to discover long-term temporal dependencies and final prediction. To confirm the effectiveness of the proposed approach, extensive experiments have been conducted on three well-known datasets i.e. UCF101, HMDB51 and YouTube. Findings shows that the resulting recognition system is on par with state-of-the-art methods.

• Journal of the HCI Society of Korea
• /
• v.2 no.2
• /
• pp.35-43
• /
• 2007

This paper presents a system for image matching and recognition based on image feature detection and description techniques from artificial satellite photographs. We propose some kind of parameters from the varied environmental elements happen by image handling process. The essential point of this experiment is analyzes that affects match rate and recognition accuracy when to change of state of each parameter. The proposed system is basically inspired by Lowe's SIFT(Scale-Invariant Transform Feature) algorithm. The descriptors extracted from local affine invariant regions are saved into database, which are defined by k-means performed on the 128-dimensional descriptor vectors on an artificial satellite photographs from Google earth. And then, a label is attached to each cluster of the feature database and acts as guidance for an appeared building's information in the scene from camera. This experiment shows the various parameters and compares the affected results by changing parameters for the process of image matching and recognition. Finally, the implementation and the experimental results for several requests are shown.

• Journal of Broadcast Engineering
• /
• v.9 no.1
• /
• pp.2-7
• /
• 2004

In this paper, we proposed the content-based music information retrieval technique using some MPEG-7 low-level descriptors. Especially, pitch information and timbral features can be applied in music genre classification, music retrieval, or QBH(Query By Humming) because these can be modeling the stochasticpattern or timbral information of music signal. In this work, we restricted the music domain as O.S.T of movie or soap opera to apply broadcasting system. That is, the user can retrievalthe information of the unknown music using only an audio clip with a few seconds extracted from video content when background music sound greeted user's ear. We proposed the audio feature set organized by MPEG-7 descriptors and distance function by vector distance or ratio computation. Thus, we observed that the feature set organized by pitch information is superior to timbral spectral feature set and IFCR(Intra-Feature Component Ratio) is better than ED(Euclidean Distance) as a vector distance function. To evaluate music recognition, k-NN is used as a classifier

• Journal of Korea Multimedia Society
• /
• v.14 no.3
• /
• pp.368-377
• /
• 2011

This paper explains the hardware structure of SURF(Speeded Up Robust Feature) based feature point extractor and its FPGA verification result. SURF algorithm produces novel scale- and rotation-invariant feature point and descriptor which can be used for object recognition, creation of panorama image, 3D Image restoration. But the feature point extraction processing takes approximately 7,200msec for VGA-resolution in embedded environment using ARM11(667Mhz) processor and 128Mbytes DDR memory, hence its real-time operation is not guaranteed. We analyzed integral image memory access pattern which is a key component of SURF algorithm to reduce memory access and memory usage to operate in c real-time. We assure feature extraction that using a Vertex-5 FPGA gives 60frame/sec of VGA image at 100Mhz.

• Journal of Internet Computing and Services
• /
• v.3 no.6
• /
• pp.13-24
• /
• 2002

This paper presents an automatic approach to detect face and facial feature from face images based on the color information and deformable model. Skin color information has been widely used for face and facial feature diction since it is effective for object recognition and has less computational burden, In this paper, we propose how to compensates varying light condition and utilize the transformed YCbCr color model to detect candidates region of face and facial feature from color images, Moreover, the detected face facial feature areas are subsequently assigned to a initial condition of active contour model to extract optimal boundaries of face and facial feature by resolving initial boundary problem when the active contour is used, The experimental results show the efficiency of the proposed method, The face and facial feature information will be used for face recognition and facial feature descriptor.

• Proceedings of the Korean Society of Broadcast Engineers Conference
• /
• 2003.11a
• /
• pp.215-218
• /
• 2003

본 논문에서는 MPEG-7에 정의된 오디오 서술자를 이용한 오디오 특징을 기반으로 한 음악 검색 알고리즘을 제안한다. 특히 timbral 특징들은 음색 구분을 용이하게 할 수 있어 음악 검색뿐만 아니라 음악 장르 분류 또는 Query by humming에 이용 될 수 있다. 이러한 연구를 통하여 오디오 신호의 대표적인 특성을 표현 할 수 있는 특징벡터를 구성 할 수 있다면 추후에 멀티모달 시스템을 이용한 검색 알고리즘에도 오디오 특징으로 이용 될 수 있을 것이다 본 논문에서는 방송 시스템에 적용 할 수 있도록 검색 범위를 특정 컨텐츠의 O.S.T 앨범으로 제한하였다. 즉, 사용자가 임의로 선택한 부분적인 오디오 클립만을 이용하여 그 컨텐츠 전체의 O.S.T 앨범 내에서 음악을 검색할 수 있도록 하였다. 오디오 특징벡터를 구성하기 위한 MPEG-7 오디오 서술자의 조합 방법을 제안하고 distance 또는 ratio 계산 방식을 통해 성능 향상을 추구하였다. 또한 reference 음악의 템플릿 구성 방식의 변화를 통해 성능 향상을 추구하였다. Classifier로 k-NN 방식을 사용하여 성능 평가를 수행한 결과 timbral spectral feature들의 비율을 이용한 IFCR(Intra-Feature Component Ratio) 방식이 Euclidean distance 방식보다 우수한 성능을 보였다.

• Proceedings of the Korean Institute of Information and Commucation Sciences Conference
• /
• 2021.05a
• /
• pp.505-507
• /
• 2021

Image recognition technology is a technology that recognizes an image object by using the generated feature descriptor and generates object feature points and feature descriptors that can compensate for the shape of the object to be recognized based on artificial intelligence technology, environmental changes around the object, and the deterioration of recognition ability by object rotation. The purpose of the present invention is to implement a power management framework required to increase profits and minimize damage to livestock farmers by preventing accidents that may occur due to the improvement of efficiency of the use of livestock house power and overloading of electricity by integrating and managing a power fire management device installed for analyzing a complex environment of power consumption and fire occurrence in a smart safety livestock house, and to develop and disseminate a safe and optimized intelligent smart safety livestock house.

• Journal of the Institute of Electronics Engineers of Korea CI
• /
• v.45 no.6
• /
• pp.170-178
• /
• 2008

A high performance object recognition algorithm using Fourier description of the 3D information of the objects is proposed. Object boundaries contain sufficient information for recognition in most of objects. However, it is not well utilized as the key solution of the object recognition since obtaining the accurate boundary information is not easy. Also, object boundaries vary highly depending on the size or orientation of object. The proposed object recognition algorithm is based on 1) the accurate object boundaries extracted from the 3D shape which is obtained by the laser scan device, and 2) reduction of the required database using the size and rotational invariant feature of the Fourier Descriptor. Such Fourier information is compared with the database and the recognition is done by selecting the best matching object. The experiments have been done on the rich database of MPEG 7 Part B.

• The Journal of the Korea Contents Association
• /
• v.17 no.3
• /
• pp.42-52
• /
• 2017

Panorama image is a single image obtained by combining images taken at several viewpoints through matching of corresponding points. Existing panoramic image generation methods that find the corresponding points are extracting local invariant feature points in each image to create descriptors and using descriptor matching algorithm. In the case of video sequence, frames may be a lot, so therefore it may costs significant amount of time to generate a panoramic image by the existing method and it may has done unnecessary calculations. In this paper, we propose a method to quickly create a single panoramic image from a video sequence. By assuming that there is no significant changes between frames of the video such as in locally, we use the FAST algorithm that has good repeatability and high-speed calculation to extract feature points and the Lucas-Kanade algorithm as each feature point to track for find the corresponding points in surrounding neighborhood instead of existing descriptor matching algorithms. When homographies are calculated for all images, homography is changed around the center image of video sequence to warp images and obtain a planar panoramic image. Finally, the spherical panoramic image is obtained by performing inverse transformation of the spherical coordinate system. The proposed method was confirmed through the experiments generating panorama image efficiently and more faster than the existing methods.

• Journal of KIISE
• /
• v.42 no.6
• /
• pp.699-706
• /
• 2015

In recent years, as near-duplicate image has been increasing explosively by the spread of Internet and image-editing technology that allows easy access to image contents, related research has been done briskly. However, BoF (Bag-of-Feature), the most frequently used method for near-duplicate image detection, can cause problems that distinguish the same features from different features or the different features from same features in the quantization process of approximating a high-level local features to low-level. Therefore, a post-verification method for BoF is required to overcome the limitation of vector quantization. In this paper, we proposed and analyzed the performance of a post-verification method for BoF, which converts SIFT (Scale Invariant Feature Transform) descriptors into 128 bits binary codes and compares binary distance regarding of a short ranked list by BoF using the codes. Through an experiment using 1500 original images, it was shown that the near-duplicate detection accuracy was improved by approximately 4% over the previous BoF method.