Earticle

정확한 인식률을 보이고 있는 상업적인 음성인식 시스템은 화자종속 고립데이터로부터 학습 모델을 사용한다. 그러 나 잡음 환경에서 데이터양에 따라 음성인식의 성능이 저하되는 문제점이 있다. 본 논문에서는 가우시안 분포에서 Maximum Log Likelihood를 이용한 벡터 양자화 기반 음성 인식 성능 향상을 제안한다. 제안하는 방법은 음성에 대한 특징 을 가지고 벡터 양자화와 Maximum Log Likelihood 음성 특징 추출 방법을 이용하여 유사 음성에 대한 음성 인식의 정확성 을 높이는 최적 학습 모델 구성 방법이다. 이를 위해 HMM을 기반으로 음성 특징을 추출하는 방법을 사용한다. 제안하는 방법을 사용하여 기존 시스템에서 생성되어 사용되는 음성 모델에 대한 부정확한 음성 모델에 대한 정확성을 향상시킬 수 있으므로 음성 인식에 강인한 모델을 구성할 수 있다. 제안하는 방법은 음성 인식 시스템에서 향상된 인식의 정확도를 보인다.

Commercialized speech recognition systems that have an accuracy recognition rates are used a learning model from a type of speaker dependent isolated data. However, it has a problem that shows a decrease in the speech recognition performance according to the quantity of data in noise environments. In this paper, we proposed the vector quantization based speech recognition performance improvement using maximum log likelihood in Gaussian distribution. The proposed method is the best learning model configuration method for increasing the accuracy of speech recognition for similar speech using the vector quantization and Maximum Log Likelihood with speech characteristic extraction method. It is used a method of extracting a speech feature based on the hidden markov model. It can improve the accuracy of inaccurate speech model for speech models been produced at the existing system with the use of the proposed system may constitute a robust model for speech recognition. The proposed method shows the improved recognition accuracy in a speech recognition system.

In this study, we propose a flexible template-matching algorithm for word segmentation, and structural analysis of features extraction is used for character recognition in the printed Arabic text. The input text image is preprocessed by the binarization and then by morphological operations. A vector quantization of the thinned image (VQTM) is created based on the idea of a freeman chain code tracking method. In the segmentation process, 113 character templates are compared for partially/completely existence in the VQTM. A non-linear filter is applied on the segmented regions to extract the termination and bifurcation features. The spatial distribution of the extracted features and other statistical characteristics are analyzed for the verification of recognition. Experimental results show that the overall recognition rate of the three fonts: Arabic transparent, simplified Arabic and traditional Arabic is 98.63%.

The performance of speaker verification system degrades when the test segments are utterances of short duration, therefore, we investigate the use of model representing our target speaker with his close speaker and his own speech data. We propose to create a new Speaker Model who groups close speakers (CS) achieved with two clustering algorithms in Automatic Speaker Verification A.S.V. Intra and Inter speaker’s variability are two clustering algorithm used in voice module. We compare the traditional approach which uses one specific customer model (Maximum a Posteriori Adaptation) with the Close Speaker model (Customers Families).Close Speaker Model (CSM) applied only when speaker model is weak achieves 42% of equal error rate. The results demonstrate that the log likelihood of close speakers is greater than the likelihood of client speaker. The false alarm from client and CSM are closest and we are constrained to enhance speaker model.

In this paper, it is considered the image compression scheme, in which a part of coded data extracted in the coding process is hidden into the other parts of coded data of own image, especially into the block address data of the best matching block within restricted blocks. The proposed scheme is able to be used in a fractal image coding in which the best matching domain block is searched, in a vector quantization in image coding in which the best matching vector is searched, and in motion compensation of moving picture in which the best matching motion vector is searched. We study each image coding method and consider the features of each coding method using the proposed scheme.

Automatic vectorization in the field of image processing and recognition is one of development direction. This paper does some relevant researches on satellite cloud image preprocessing, image segmentation, image vector quantization. To improve the effect of noise reduction and preserving image details, this paper puts forward an improved adaptive median filter algorithm; For increasing the speed of image segmentation, this paper puts forward automatic layering algorithm combined with color information. Finally, this paper puts forward automatic vector quantization algorithm based on satellite cloud images and we developed an automatic vector quantization prototype system of satellite cloud images. The research results suggest that our automatic vector quantization algorithm has satellite cloud information automatic extraction function, identification function and vector quantization function.

A method for image compression, adaptive difference compensation vector quantization using dynamic image block adjustment is proposed in this paper. Before image coding, this proposed method analyzes the similarity values between the encoding image sub-block and its 8-neighbor sub-blocks, then determines the encoding rule according to the similarity value and the preset threshold. If the similarity value is approximate to the threshold, we use the same codeword to encode the neighbor block and the encoding image block; otherwise, the neighbor block is separately encode. When encoding, the difference between each image block and its matching codeword is first computed to obtain the difference image, and then the sign bits of the pixel difference is imposed the running length coding on and attached after the codeword index. When decoding, this proposed method restores the compressed image according to the codeword index, performs the running length encoding to deal with decode the attached information, uses the window of 33 size to tackle the adaptive difference compensation and derive the final decoding image. The experiment results reveal that this proposed method can improve the encoding speed and image restoration performance against the normal vector quantization.

The mean-removed vector quantization (MRVQ) scheme achieves good reconstructed image quality, but it requires a high number of bit rates. In this paper, we propose an improved MRVQ scheme. The compressed codes of MRVQ for an image block contain the block mean and the index recording the closest residual vector in the codebook. In the proposed scheme, the block mean values are encoded by the linear prediction technique followed by the Huffman coding technique. The MRVQ indices of the residual vectors are further compressed by the Huffman coding technique. From the experimental results, it is shown that a great deal of bit rate reduction is achieved by using the proposed scheme with acceptable image quality loss.

As the digital information must be stored and recovered in an effective ways, it is necessary to compress data in image coding, wavelet transform and lattice vector quantization are adopted in this paper to improve the compression effective of image coding. We first propose the two dimensional spatial wavelet decomposition of image, and then we present lattice Vector Quantization coding scheme for each image subband, lattice vector quantization can make full use of the correlation between the image wavelet coefficients to remove the information redundancy, and the coding method of lattice quantization can be effective because of applying the symmetries of the lattice. The establishment of rate distortion (RD) model suitable for lattice vector quantization of wavelet image coder is also important for image compression, study shows that the RD performance for the spatial subbands are fitted by an exponential form theoretical model, this yields an analytical solution to the bit rate distribution issue, we explore an effective rate control scheme by using the lagrangian optimization method to distribute the bit rate for the spatial subbands. The experimental results show that the proposed algorithm can achieve better compressing effect with minimum loss.

In this paper, an image compression scheme is proposed, based on discrete cosine transform (DCT). This scheme is a hybrid method, which combines vector quantization (VQ) and differential pulse code modulation (DPCM). This scheme begins with transforming image from spatial domain to frequency domain using DCT. Then the block data is transformed into a vector according to zigzag order, and then truncated. After that, the vector is split into DC coefficient and AC coefficients. After scale quantization, DC coefficient is coded using DPCM. AC coefficients are coded using multistage vector quantization (MSVQ). Then, entropy encoding is performed on index-tables and DC part, separately. The experimental results show that, compared to conventional VQ and DCT-VQ schemes, proposed scheme has a better performance.

The Session Initiation Protocol (SIP) is the main protocol behind Voice over IP (VoIP). However, it does not provide authentication, which may lead to possible impersonation and eavesdropping threats. Negotiating keys using digital certificates may help secure the channel, but this method incurs extra maintenance cost. Verbal authentication utilizes the real-time nature of VoIP but it requires users to read and manually compare the authentication code. This study proposes a session identifier framework based on the Modified Vector Quantization (MVQ) method on real-time video frames for video communication in VoIP. After the selected image frames are averaged and quantized, the output of the MVQ process is a set of image metrics that serves as pre-shared keys for key agreement. The framework is certificate-less and users do not need to read the authentication code to the other user. The implementation is evaluated for accuracy and robustness towards network noise and frame conditions.

본 논문에서는 기계학습의 하나인 Learning Vector Quantization 알고리즘을 이용하여 컨트롤러 순서를 정하는 모델을 제안하였다. 제안한 모델은 모든 컨트롤러 정보를 수집하여 Learning Vector Quantization의 LVQ1와 LVQ2 기법을 이용하여 컨트롤러의 순서를 정한다. 이를 통해, 효율적인 컨트롤러 동기화가 이뤄질 것으로 기대된다.

Compression of magnetic resonance images (MRI) has proved to be more difficult than other medical imaging modalities. In an average sized hospital, many tora bytes of digital imaging data (MRI) are generated every year, almost all of which has to be kept. The medical image compression is currently being performed by using different algorithms. In this paper, Fuzzy C-Means (FCM) algorithm is used for the Vector Quantization (VQ). First, a digital image is divided into subblocks of fixed size, which consists of 4${\times}$4 blocks of pixels. By performing 2-D Discrete Cosine Transform (DCT), we select six DCT coefficients to form the feature vector. And using FCM algorithm in constructing the VQ codebook. By doing so, the algorithm can make good time quality, and reduce the processing time while constructing the VQ codebook.

In this paper, we have proposed a VQ algorithm which uses a generating order to make quantize feature vector of speech signal. The proposed algorithm inspects what codeword follows a(ter present codeword and adds new index to established codebook, when mapping speech signal. We present a variable bit rate for new codebook, and propose an efficient compressed way of information. In this way, the number of computation and the number of codewords to be searched are reduced considerably. The performance of the proposed VQ algorithm is evaluated by spectrum distortion measure and bit rate. The obtained spectrum distortion is reduced about 0.22 [db], and the bit rate is saved over 0.21 bit/frame.

In this paper, a new competition strategy for learning vector quantization is proposed. The simple competitive strategy used for learning vector quantization moves the winning prototype which is the closest to the newly given data pattern. We propose a new learning strategy based on k-nearest neighbor prototypes as the winning prototypes. The selection of several prototypes as the winning prototypes guarantees that the updating process occurs more frequently. The design is illustrated with the aid of numeric examples that provide a detailed insight into the performance of the proposed learning strategy.

본 논문에서는 LVQ를 퍼지화한 새로운 퍼지 학습 법칙들을 제안하였다. 퍼지 LVQ 학습법칙 1은 기존의 학습률 대신에 퍼지 학습률을 사용하였는데 이는 조건 확률의 퍼지화에 기반을 두고 있다. 퍼지 LVQ 학습법칙 2는 클래스들 사이에 존재하는 입력벡터가 결정 경계선에 대한 정보를 더 가지고 있는 것을 반영한 것이다. 이 새로운 퍼지 학습 법칙들을 improved IAFC(Integrted Adaptive Fuzzy Clustering)신경회로망에 적용하였다. improved IAFC신경회로망은 ART-1 (Adaptive Resonance Theory)신경회로망과 Kohonen의 Self-Organizing Feature Map의 장점을 취합한 퍼지 신경회로망이다. 제안한 supervised IAFC 신경회로망 1과 supervised IAFC neural 신경회로망 2의 성능을 오류 역전파 신경회로망의 성능과 비교하기 위하여 iris 데이터를 사용하였는데 Supervised IAFC neural network 2가 오류 역전파 신경회로망보다 성능이 우수함을 보여주었다.

In the compression methods widely used today, the image compression by VQ is the most popular and shows a good data compression ratio. Almost all the methods by VQ use the LBG algorithm that reads the entire image several times and moves code vectors into optimal position in each step. This complexity of algorithm requires considerable amount of time to execute. To overcome this time consuming constraint, we propose an enhanced self-organizing neural network for color images. VQ is an image coding technique that shows high data compression ratio. In this study, we improved the competitive learning method by employing three methods for the generation of codebook. The results demonstrated that compression ratio by the proposed method was improved to a greater degree compared to the SOM in neural networks.

본 연구에서는 퍼지사상화(fuzzy mapping)에 의한 사상된(mapped) 코드북을 사용하는 화자적은 음성합성 알고리즘을 제안한다. 입력화자와 기준화자의 코드북은 신경망 클러스터링 알고리즘인 자율경쟁 학습을 사용하여 작성된다. 사상된 코드북은 입력 음성벡터에 대한 두 화자의 대응 코드벡터의 소속갑(membership value)으로 퍼지 히스토그랩을 작성하여 이들을 1차 결합함으로써 얻어지는 퍼지사상화에 의하여 작성된다. 음성합성시에는 사상된 코드북을 사용하여 입력화자의 음것을 퍼지 벡터양자화한 다음, CFM 연산으로 합성함으로써 입력화자에 적응된 합성음을 얻는다. 실험에서 여러 입력화자로 30대의 남성, 20대의 여성음을 사용하였고 기준음석으로 입력음성과는 다른 20대의 여성음성을 사용하였다.실험에 사용된 음성데이타는 문장/안녕하십니까/와/굿모닝/이다. 실험결과는 각각의 입력화자에 기준화자 음성이 적응된 합성음을 얻었다.

Compressing image data by using Vector Quantization (VQ)［1］-［3］will compare Training Vectors with Codebook. The result is an index of position with minimum distortion. The implementing Random Codebook will reduce the image quality. This research presents the Splitting solution ［4］,［5］to implement the Codebook, which improves the image quality［6］by the average Training Vectors, then splits the average result to Codebook that has minimum distortion. The result from this presentation will give the better quality of the image than using Random Codebook.

In this paper, we propose an effective method for music summarization which automatically extracts a representative part of the music by using signal processing technology. Proposed method uses a vector quantization technique to extract several segments which can be regarded as the most important contents in the music. In general, there is a repetitive pattern in music, and human usually recognizes the most important or catchy tune from the repetitive pattern. Thus the repetition which is extracted using segment similarity is considered to express a music summary. The segments extracted are again combined to generate a complete music summary. Experiments show the proposed method captures the main theme of the music more effectively than conventional methods. The experimental results also show that the proposed method could be used for real-time application since the processing time in generating music summary is much faster than other methods.