• The Journal of the Acoustical Society of Korea
• /
• v.6 no.3
• /
• pp.60-67
• /
• 1987

In this paper, a phoneme classification method for Korean speech recognition has been proposed and its performance has been studied. The phoneme classification has been done based on the phonemic features extracted by the prewindowed recursive least-square (PRLS) algorithm that is a kind of adaptive filter algorithms. Applying the PRLS algorithm to input speech signal, precise detection of phoneme boundaries has been made, Reference patterns of Korean phonemes have been generated by the ordinery vector quantization (VQ) of feature vectors obtained manualy from prototype regions of each phoneme. In order to obtain the performance of the proposed phoneme classification method, the method has been tested using spoken names of seven Korean cities which have eleven different consonants and eight different vowels. In the speaker-dependent phoneme classification, the accuracy is about $85\%$ considering simple phonemic rules of Korean language, while the accuracy of the speaker-independent case is far less than that of the speaker-dependent case.

• The Journal of Information Technology
• /
• v.7 no.4
• /
• pp.1-8
• /
• 2004

The depth information of 3D image lost by projecting 3D-object to 2D-screen for earning image. If depth information is restored and is used to recognize 3D-object, we can make the more effective recognition system. We often use shape from stereo algorithm in order to restore this information. In this paper, we suggest 3-D object recognition system in which the 3-D Hough transform domain is employed to represent the 3-D objects. In this system, we use the moving vector of object to reduce matching time and In second matching step, the unknown input image is compared with the reference images, which is made with octree codes. Octree codes are used in volume-based representation of a three dimensional object. The result of simulation show that the proposed 3-D object recognition system provides satisfactory performance.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.9 no.7
• /
• pp.2599-2613
• /
• 2015

Multi-index hashing (MIH) is the state-of-the-art method for indexing binary codes, as it di-vides long codes into substrings and builds multiple hash tables. However, MIH is based on the dataset codes uniform distribution assumption, and will lose efficiency in dealing with non-uniformly distributed codes. Besides, there are lots of results sharing the same Hamming distance to a query, which makes the distance measure ambiguous. In this paper, we propose a data-oriented multi-index hashing method (DOMIH). We first compute the covariance ma-trix of bits and learn adaptive projection vector for each binary substring. Instead of using substrings as direct indices into hash tables, we project them with corresponding projection vectors to generate new indices. With adaptive projection, the indices in each hash table are near uniformly distributed. Then with covariance matrix, we propose a ranking method for the binary codes. By assigning different bit-level weights to different bits, the returned bina-ry codes are ranked at a finer-grained binary code level. Experiments conducted on reference large scale datasets show that compared to MIH the time performance of DOMIH can be improved by 36.9%-87.4%, and the search accuracy can be improved by 22.2%. To pinpoint the potential of DOMIH, we further use near-duplicate image retrieval as examples to show the applications and the good performance of our method.

• The Journal of the Korea Contents Association
• /
• v.16 no.6
• /
• pp.9-18
• /
• 2016

This paper proposes a multi-view Wyner-Ziv Video coding scheme based on spatio-temporal adaptive estimation. The proposed algorithm is designed to search for a better estimated block with joint bi-directional motion estimation by introducing weights between temporal and spatial directions, and by classifying effectively the region of interest blocks, which is based on the edge detection and the synthesis, and by selecting the reference estimation block from the effective motion vector analysis. The proposed algorithm exploits the information of a single frame viewpoint and adjacent frame viewpoints, simultaneously and then generates adaptively side information in a variety of closure, and reflection regions to have a better performance. Through several simulations with multi-view video sequences, it is shown that the proposed algorithm performs visual quality improvement as well as bit-rate reduction, compared to the conventional methods.

• Journal of Korea Multimedia Society
• /
• v.13 no.2
• /
• pp.185-195
• /
• 2010

In this paper, we propose an efficient inter and intra prediction algorithms for fast encoding of H.264/AVC. First, inter prediction mode decision method decides early using temporal/spatial correlation information and pixel direction information. Second, intra prediction mode decision method selects block size judging smoothness degree with inner/outer pixel value variation and decides prediction mode using representative pixel and reference pixel. Lastly, adaptive motion search region restriction sets search region using mode information of neighboring block and predicted motion vector. The experimental results show that proposed method can achieve about 18~53% reduction compared with the existing JM 14.1 in the encoding time. In RD performance, the proposed method does not cause significant PSNR value losses while increasing bitrates slightly.

• Journal of Korea Multimedia Society
• /
• v.13 no.7
• /
• pp.991-999
• /
• 2010

In this paper, we propose a fast motion estimation algorithm for video encoding. Conventional fast motion estimation algorithms have a serious problem of low prediction quality in some frames. However, full search based fast algorithms have low computational reduction ratio. In the paper, we propose an algorithm that significantly reduces unnecessary computations, while keeping prediction quality almost similar to that of the full search. The proposed algorithm uses distribution probability of motion vectors and adaptive search patterns and block matching criteria. By taking different search patterns and error criteria of block matching according to distribution probability of motion vectors, we can reduces only unnecessary computations efficiently. Our algorithm takes only 20~30% in computational amount and has decreased prediction quality about 0~0.02dB compared with the fast full search of the H.264 reference software. Our algorithm will be useful to real-time video coding applications using MPEG-2 or MPEG-4 AVC standards.

• The Transactions of the Korean Institute of Power Electronics
• /
• v.8 no.4
• /
• pp.321-327
• /
• 2003

As the continuous casting process is to product slab with high temperature liquid steel, the main role of strand driven roll is to withdraw slab from mold as operator set up casting speed pattern. The strand driven roll in old cast machine is controlled casting speed only. Due to inaccuracies in drive setting up, varying roll diameters, bulging in the product, withdrawal force was distributed irregularly. As a result, because of horizontal crack in slab comer, high casting speed can't be achieved. In this paper, the correlation between the distribution of withdrawal force and slab quality is investigated and the new control algorithm which can be distributed regularly the withdrawal force of strand driven roll is proposed. The principle of proposed algorithm is not to control motor torque directly but to control motor speed reference according to sharing ratio of withdrawal force which is set up in high level controller. The proposed algorithm implemented in POSCO Kwangyang 1-4 continuous casting plant.

• Proceedings of the KSRS Conference
• /
• 2004.10a
• /
• pp.626-629
• /
• 2004

Texture imaging, which means texture image creation by co-occurrence relation, has been known as one of useful image analysis methodologies. For this purpose, most commercial remote sensing software provides texture analysis function named GLCM (Grey Level Co-occurrence Matrix). In this study, texture-imaging program for GLCM algorithm is newly implemented in the MS Visual IDE environment. While, additional texture imaging modules based on GLDV (Grey Level Difference Vector) are contained in this program. As for GLCM/GLDV texture variables, it composed of six types of second order texture function in the several quantization levels of 2(binary image), 8, and 16: Homogeneity, Dissimilarity, Energy, Entropy, Angular Second Moment, and Contrast. As for co-occurrence directionality, four directions are provided as $E-W(0^{\circ}),\;N-E(45^{\circ}),\;S-W(135^{\circ}),\;and\;N-S(90^{\circ}),$ and W-E direction is also considered in the negative direction of E- W direction. While, two direction modes are provided in this program: Omni-mode and Circular mode. Omni-mode is to compute all direction to avoid directionality problem, and circular direction is to compute texture variables by circular direction surrounding target pixel. At the second phase of this study, some examples with artificial image and actual satellite imagery are carried out to demonstrate effectiveness of texture imaging or to help texture image interpretation. As the reference, most previous studies related to texture image analysis have been used for the classification purpose, but this study aims at the creation and general uses of texture image for urban remote sensing.

• Journal of the Korea Society of Computer and Information
• /
• v.10 no.5 s.37
• /
• pp.77-86
• /
• 2005

In this paper, we designed a learning algorithm of LVQ that extracts classification errors and learns ones and improves classification performance. The proposed LVQ learning algorithm is the learning Networks which is use SOM to learn initial reference vectors and out-star learning algorithm to determine the class of the output neurons of LVQ. To extract pattern vectors which cause classification errors, we proposed the error-cause condition, which uses that condition and constructed the pattern vector space which consists of the input pattern vectors that cause the classification errors and learned these pattern vectors , and improved performance of the pattern classification. To prove the performance of the proposed learning algorithm, the simulation is performed by using training vectors and test vectors that are Fisher' Iris data and EMG data, and classification performance of the proposed learning method is compared with ones of the conventional LVQ, and it was a confirmation that the proposed learning method is more successful classification than the conventional classification.

• Journal of the Korean Institute of Intelligent Systems
• /
• v.15 no.1
• /
• pp.12-17
• /
• 2005

The segment matching method shows better performance than the global and points-based methods to compare reference signature with an input signature. However, the segment-to-segment matching method has the problem of decreasing recognition rate according to the variation of partitioning points. This paper proposes a fusion model based on the segment matching and HMM to construct a more reliable authentic system. First, a segment matching classifier is designed by conventional technique to calculate matching values lot dynamic information of signatures. And also, a novel HMM classifier is constructed by using the principal component analysis to calculate matching values for static information of signatures. Finally, SVM classifier is adopted to effectively combine two independent classifiers. From the various experiments, we find that the proposed method shows better performance than the conventional segment matching method.