• IEMEK Journal of Embedded Systems and Applications
• /
• v.15 no.1
• /
• pp.1-8
• /
• 2020

A derivation method for a quantized gradient for machine learning on an embedded system is proposed, in this paper. The proposed differentiation method induces the quantized gradient vector to an objective function and provides that the validation of the directional derivation. Moreover, mathematical analysis shows that the sequence yielded by the learning equation based on the proposed quantization converges to the optimal point of the quantized objective function when the quantized parameter is sufficiently large. The simulation result shows that the optimization solver based on the proposed quantized method represents sufficient performance in comparison to the conventional method based on the floating-point system.

• Journal of information and communication convergence engineering
• /
• v.16 no.3
• /
• pp.142-147
• /
• 2018

We present a practical design algorithm for quantizers at nodes in distributed systems in which each local measurement is quantized without communication between nodes and transmitted to a fusion node that conducts estimation of the parameter of interest. The benefits of vector quantization (VQ) motivate us to incorporate the VQ strategy into our design and we propose a low-complexity design technique that seeks to assign vector codewords into sets such that each codeword in the sets should be closest to its associated local codeword. In doing so, we introduce new distance metrics to measure the distance between vector codewords and local ones and construct the sets of vector codewords at each node to minimize the average distance, resulting in an efficient and independent encoding of the vector codewords. Through extensive experiments, we show that the proposed algorithm can maintain comparable performance with a substantially reduced design complexity.

• Speech Sciences
• /
• v.13 no.4
• /
• pp.177-186
• /
• 2006

In an effort to enhance the quality of feature vector classification and thereby reduce the recognition error rate of the speaker-independent speech recognition, we employ the Mahalanobis distance in the calculation of the similarity measure between feature vectors. It is assumed that the metric matrix of the Mahalanobis distance be diagonal for the sake of cost reduction in memory and time of calculation. We propose that the diagonal elements be given in terms of the variations of the feature vector components. Geometrically, this prescription tends to redistribute the set of data in the shape of a hypersphere in the feature vector space. The idea is applied to the speech recognition by hidden Markov model with fuzzy vector quantization. The result shows that the recognition is improved by an appropriate choice of the relevant adjustable parameter. The Viterbi score difference of the two winners in the recognition test shows that the general behavior is in accord with that of the recognition error rate.

• The Journal of Korean Institute of Communications and Information Sciences
• /
• v.37A no.12
• /
• pp.1043-1053
• /
• 2012

In distributed video coding (DVC) systems, the complexity of encoders is greatly reduced by removing the motion estimation operations in encoders, since the correlation between frames is utilized in decoders. The transmission of parity bits is requested through the feedback channel, until the related errors are corrected to decode the Wyner-Ziv frames. The requirement to use the feedback channel limits the application of DVC systems. In this paper, we propose an efficient method to remove the feedback channel in DVC systems. First, a simple side information generation method is proposed to calculate the amount of parity bits in the encoder, and it is shown that the proposed method yields good performance with low complexity. Then, by calibrating the theoretical entropy with three parameters, we can calculate the amount of parity bits in the encoder and remove the feedback channel. Moreover, an adaptive method to determine quantization parameters for key frames is proposed. Extensive computer simulations show that the proposed method yields better performance than conventional methods.

• Journal of Korea Multimedia Society
• /
• v.9 no.7
• /
• pp.818-826
• /
• 2006

In this paper, we investigate the relationship of bitrate and quantization parameter needed for the trans-coding method that makes the H.264 bitstream of a particular bitrate to the other titrate. Also we propose the new method in order to transcode the titrate between H.264 video coded bitstreams. The proposed transcoding method updates the model parameters from previous picture or slice by using the approximated relationship of bitrate and quantization step-size and finds the target quantization step-size, and then generates the target titrate by simple coding processing just after requantization. Therefore, the proposed method does not need the complex bitrate control and converts to the target titrate by simple implementation. From simulation, we can see that the proposed method transcodes exactly to an assigned target bitrate for the four test sequences with different their characteristics.

• Journal of IKEEE
• /
• v.5 no.2 s.9
• /
• pp.221-231
• /
• 2001

In this paper, we propose an efficient requantization method for INTRA frames in heterogeneous transcoding from MPEG-1 to MPEG-4 simple profile. The quantizer for MPEG-1 INTRA MB usually uses a quantization weighting matrix while the quantizer for MPEG-4 simple profile doesn't. As a result, the quantization step sizes of the two quantizers may not be the same even for the same quantization parameter. Due to this mismatch in the quantization step site, the transcoded MPEG-4 sequence suffers from serious quality degradation and the number of bits produced by transcoding increases from the original MPEG-1 video sequence. To solve these problems, we propose an efficient method to find a near-optimum reconstruction level in the transcoder. We also present a PDF (probability distribution function) estimation method for the original DCT coefficients of MPEG-1 video sequence, which is required for the proposed requantization. Experimental results show that the proposed method gives $0.3{\sim}0.6dB$ improvement in PSNR over the conventional method, even at the reduced bit-rate about $5{\sim}7%$ from the conventional method.

• The Journal of the Institute of Internet, Broadcasting and Communication
• /
• v.21 no.1
• /
• pp.93-98
• /
• 2021

Due to the development of deep learning and AI, the scale of the model has grown, and it has been integrated into other fields to blend into our lives. However, in environments with limited resources such as embedded devices, it is exist difficult to apply the model and problems such as power shortages. To solve this, lightweight methods such as clouding or offloading technologies, reducing the number of parameters in the model, or optimising calculations are proposed. In this paper, quantization of learned models is applied to ONNX models used in various framework interchange formats, neural network structure and inference performance are compared with existing models, and various module methods for quantization are analyzed. Experiments show that the size of weight parameter is compressed and the inference time is more optimized than before compared to the original model.

• Journal of KIISE:Information Networking
• /
• v.35 no.3
• /
• pp.235-242
• /
• 2008

To improve video quality and coding efficiency, H.264/AVC adopted an adaptive rate control. But this method has a problem as it cannot predict an accurate quantization parameter(QP) for the first frame. The first QP is decided among four constant values by using encoder input parameters. It does not consider encoding bits, results in significant fluctuation of the image quality and decreases the average quality of the whole coded sequence. In this paper, we propose a new algorithm for the first frame QP decision in the H.264/AVC encoder. The QP is decided by the existing algorithm and the first frame is encoded. According to the encoded bits, the new initial QP is decided. We can predict optimal value because there is a linear relationship between encoded bits and the new initial QP. Next, we re-encode the first frame using the new initial QP. Experimental results show that the proposed algorithm not only achieves better quality than the state of the art algorithm, but also adopts a rate control forthe sequence that was impossible with the existing algorithm. By reducing fluctuation, subjective quality also improved.

• IEIE Transactions on Smart Processing and Computing
• /
• v.6 no.4
• /
• pp.276-280
• /
• 2017

This paper presents computationally efficient image compression for wireless transmission of high-definition video, to adaptively utilize available channel bandwidth and improve image quality. The method indirectly predicts an unknown available channel bandwidth by monitoring encoder buffer status, and adaptively controls a quantization parameter to fully utilize the bandwidth. Experimental results show that the proposed method is robust to variations in channel bandwidth.

• The Journal of the Acoustical Society of Korea
• /
• v.15 no.1E
• /
• pp.34-38
• /
• 1996

In this paper, a new fast voronoi divider for vector quantization (VQ) is introduced, which results from Theorem that the nearest vectors in the sense of minimum mean square error(MMSE) have almost the same mean values of their elements. An improved splitting method for a VQ codebook design using the fast voronoi divider is also presented. Experimental results show that the new method reduces the complexity of training a VQ codebook several times with a high signal to noise ratio(SNR) using an appropriate extensive parameter of codebook.