Kwang-Deok Seo

Robust DTW-based recognition algorithm for hand-held consumer devices

This work presents a new dynamic programming (DP)-based recognition algorithm that is quite suitable for menu-driven recognition applications with small vocabulary size (typically less than 50). In terms of computational complexity and parameter file size, the proposed algorithm shows improved results for a command recognition system with small vocabulary size than the conventional hidden Markov model (HMM)-based recognizer. In addition, the proposed algorithm shows much higher recognition accuracy compared to the conventional dynamic time warping (DTW)-based recognizer.

Wipe scene-change detector based on visual rhythm spectrum

To store and retrieve large-scale video data sets effectively, the process of shot-change detection is an essential step. In this paper, we propose an automatic shot-change detection algorithm based on Visual Rhythm Spectrum. The Visual Rhythm Spectrum contains distinctive patterns or visual features for many different types of video effects. For the improvement of detection speed, the proposed algorithm is executed by using the partial data of digital compressed video. The proposed detection algorithm can be universally applied to various kinds of shot-change categories such as scene-cuts and wipes. The developed wipe detector is implemented and tested with real video sequences containing a variety of wipe types and lengths. It is shown by simulations that the proposed detection algorithm outperforms other existing approaches.

Fast motion vector re-estimation for transcoding MPEG-1 into MPEG-4 with lower spatial resolution in DCT-domain

In this paper, we propose a fast motion vector re-estimation for transcoding MPEG-1 to MPEG-4 with lower spatial resolution. This task can be performed in the pixel-domain or in the discrete cosine transform (DCT) domain. In this paper, we concentrate on the DCT-domain approach, which requires lower delay and complexity than those in the pixel-domain. For the DCT-domain transcoding to lower spatial resolution pictures, DCT-domain down-sampling filter is applied and a base motion vector (BMV) for the down-sampled MPEG-4 macroblock is to be calculated from the input motion vectors operating on the higher spatial resolution image. Quality can be significantly improved by refining the BMV. Starting with the BMV, the motion vector refinement (MVR) scheme searches for a delta motion vector within a significantly reduced search area. We propose a fast MVR scheme for video down-sampling in the DCT-domain based on minimizing the number of required check points, and a computationally efficient method for extracting motion compensated DCT block. We also show an efficient scheme for selecting a coding mode for a macroblock in the lower resolution video from those of the corresponding higher resolution video.

Data partitioning and coding of DCT coefficients based on requantization for error-resilient transmission of video

In this paper, we propose an efficient data partitioning and coding algorithm for an error-resilient transmission of DCT coefficients in error prone environment. In the typical data partitioning for Inter-coded frames, motion and macroblock header information is separated from the texture information. It can be an effective tool for the transmission of video over the error prone environment. For Intra-coded frames, however, the loss of DCT coefficients is fatal because there is no other information to reconstruct the corrupted macroblocks by errors. Conventional data partitioning algorithm for DCT coefficients is to separate a fixed number of the significant DCT coefficients from the remaining coefficients, called the spectral separation. While the spectral separation can guarantee an error resilient transmission with small overhead, the main drawback is a significant decrease in the image quality of the high priority partition, compared with that of the bitstreams without data partitioning for an equivalent bit-rate. In the proposed scheme, the quantized DCT coefficients are partitioned into an even-value approximation and the odd remainder part. We also propose a simple and efficient coding algorithm for the odd remainder part. It is shown that the proposed algorithm provides a better image quality than the conventional methods with a little overhead.

A robust formant extraction algorithm combining spectral peak picking and root polishing

We propose a robust formant extraction algorithm that combines the spectral peak picking, formants location examining for peak merger checking, and the root extraction methods. The spectral peak picking method is employed to locate the formant candidates, and the root extraction is used for solving the peak merger problem. The location and the distance between the extracted formants are also utilized to efficiently find out suspected peak mergers. The proposed algorithm does not require much computation, and is shown to be superior to previous formant extraction algorithms through extensive tests using TIMIT speech database.

Dynamic bit-rate reduction based on frame-skipping and requantization for MPEG-1 to MPEG-4 transcoder

In this paper, we propose a dynamic bit-rate reduction scheme for transcoding an MPEG-1 bitstream into an MPEG-4 simple profile bitstream. For dynamic bit-rate reduction, a significant reduction in the bit-rate is achieved by combining the processes of frame-skipping and requantization. For bit-rate reduction by frame-skipping, an efficient method is proposed for estimating the relevant motion vectors from the skipped frames. The conventional FDVS (forward dominant vector selection) method is improved to reflect the effect of the macroblock types in the skipped frames. For bit-rate reduction by requantization, an efficient requantization method is used for an MPEG-1 to MPEG-4 transcoder, consisting of R-Q (rate-quantization) modeling with a simple feedback and an adjustment of the quantization parameters to compensate for the different coding efficiency between MPEG-1 and MPEG-4. Simulation results demonstrated that the proposed method combining frame-skipping and requantization can generate a transcoded MPEG-4 bitstream that is much closer to the desired low bit-rate than the conventional method along with a superior visual quality.

A new timing model design for MPEG Media Transport (MMT)

During the recent years, broadcasting services and mobile services have started converging and it is expected that this trend will continue with other services. In order to deploy efficient solutions for the transport of MPEG media in an interoperable fashion, especially given the recent increased demand of multimedia delivery in the heterogeneous network environment, MPEG has launched a new standardization work item, called MPEG Media Transport (MMT) since the middle of 2010. In this paper, we propose an appropriate timing model for MPEG Media Transport (MMT) to support media synchronization in MMT system based media service.

A fast video mixing method for multiparty video conference

In this paper, we propose a fast video mixing method for reducing the computational complexity in the MCU (Multipoint Control Unit) used in the video conferencing. The conventional mixing method is based on the pixel-domain transcoder, of which computational complexity is linearly increased as the number of participants is increased. Basically the method requires many decoders and one huge encoder to mix the multiple bitstreams. To reduce the computational complexity, we propose a hybrid mixing method based on the syntax-based bitstream modification and pixel-domain transcoder. The proposed method reduces the computational complexity about 45% at the improved quality, compared with the conventional mixing method based on the pixel-domain transcoder.

Adaptive rate control algorithm based on logarithmic R–Q model for MPEG-1 to MPEG-4 transcoding

Abstract This paper proposes an adaptive rate control algorithm for MPEG-1 to MPEG-4 transcoder for video internetworking where a compressed video is transmitted through networks of different capacities. It is based on an adaptive logarithmic R–Q (rate-quantization) model for controlling the new target bit-rate of a transcoded MPEG-4 bitstream. In R–Q modeling, the R–Q characteristics predicted from a previous picture of the same picture type are used in combination with the R–Q data measured for the current picture. The proposed adaptive model takes into account the effect of picture dependencies arising in motion-compensated video coding. Furthermore, full advantage is taken of the special situation related to transcoding where a priori knowledge of the current R–Q data can be obtained by just counting the number of bits in the input bitstream without re-encoding. It is shown by simulation that the adaptive R–Q model-based rate control can efficiently control the output bit-rate to match the target bit-budget.

A client-driven media synchronization mechanism for RTP packet-based video streaming

AbstractMedia synchronization is used to correctly playback a video stream with its associated audio. To support synchronization between video and audio streams transported over IP networks, an RTP/RTCP protocol suite is usually employed. In conventional server-driven media synchronization, the server needs to periodically transmit an RTCP sender report (SR) packet to provide the client with a UTC time in NTP format corresponding to the RTP timestamp carried by each RTP packet. In this paper, we propose a precise client-driven media synchronization mechanism for an RTP packet-based multimedia streaming service. In the proposed method, the server does not need to send any RTCP SR packets for synchronization. Instead, the client device derives the precise normal play time (NPT) for each video and audio stream from the received RTP packets containing an RTP timestamp. Simulations show that the proposed client-driven synchronization method can provide accurate media synchronization without employing an RTCP SR packet and accordingly reduce the initial synchronization delay, the processing complexity at the client device, the number of required user datagram protocol ports, and the amount of control traffic injected into the network.