The previously developed universal pattern-matching coding technique (PMC) uses different coding modes, and adaptively chooses the proper coding mode instantaneously for a given input subblock. To improve the adaptability of the coding-mode selection, we introduced the delayed-decision mechanism with the (M,L) algorithm, which can be considered a kind of multi-path tree search algorithm, and reported its superiority over the existing instantaneous coding-mode selection. However, the coding performance of our previousely developed method degraded because of the discontinuity of raster scanning. The work herein proposes a scanning which does not contain any discontinuity and demonstrates that this scanning method improves coding performance to some extent.

View full abstract

When the speech spoken by a person on the screen is translated into the one spoken in a different language, the mismatching between the speech and the mouth shapes will be observed. This type of problem also occurs in the multilingual broadcasting of TV programs. This paper proposes a novel method to reduce the mismatching between the speech spoken in a different language and the mouth shapes in the moving facial images. Two basic techniques are utilized to carry out the conversion of mouth shapes. One is the rule-based conversion from text to mouth shapes and speech. The other is the 3-D motion estimation of whole head using the distribution of 2-D motion vectors detected from the sequence of input frames and a 3-D-shape model of head. After obtaining the converted mouth shapes represented by the 3-D-shape model, the original texture of mouth area is mapped onto the 3-D-shape model. Then, this mouth image replaces the mouth area in the input frame. Some experimental results are shown to demonstrate the effect and usefulness of the proposed method.

View full abstract

This paper addresses a reconstruction of a image signal with zero-crossings and additional information based on dyadic wavelet transform, and treats the application to the noise reduction. At first, for 1-dimensional signal, the reconstruction algorithm is described. The zero-crossing points are detected by using the multiscale dyadic wavelet trnsform. The signal is reconstructed with the iteration method of the projection onto convex sets. The convex sets used are obtained from the wavelet transform coefficients space and the other space. The other space is constructed with the constraints for the wavelet transform of the given input signal. This reconstruction method is extended to the 2-dimensional signal. Finally, as an application of the algorithm, the noise reduction method based on the appropriate threshold values is presented.

View full abstract

[in Japanese]

View full abstract

The hierarchical coding technique is useful for randam access and error concealment, because it is possible to extract part of VLC. But it may have two problems: deterioration of coding efficiency by hierarchical structure, and the difficulty of rate control on each layer. In this paper, the hierarchical structure is built on DCT coefficients. To establish the better hierarchical system, the minimum required bits on Lower frequency Layer is considered, and the techinical studies on rate control and high picture quality method are discussed.

View full abstract

In this paper, we propose an advanced scheme of arithmetic coding based upon the adaptive prediction Markov model for grey scale images, assuming the probability. distribution of the image signal set. Furthermore, we present a method to determine the initial values of the approximated probability distribution's table and a method of updating the probability table by estimating the parameters of the assumed probability distribution inferenced through the learn-ing of the image single set. In addition, we evaluate these approaches by computer simulation.

View full abstract

MPEG2 interframe prediction written in "Test Model 2 (TM2)" is introduced. "Test Model 2" is a document describing a coding algorithm and a bit stream syntax for simulations to develop MPEG2 standard. Source images in MPEG2 are interlaced signals. Thus, an effective prediction that can handle the interlaced signal is considered. It is mainly based on field/frame adaptive prediction in the macroblock layer and picture structure in the picture layer. Other special predictions are also described in this paper.

View full abstract

The quantization and coding control in MPEG-2 standardization effort are described, in which the quantization scheme of MPEG-1 is as the contrast to H.261 and JPEG, scanning and two-dimensional VLC are in preparation for the MPEG second phase coding problems. Rate control methods in the SM3 and TM's are described with some non-trivial detected defects. The high quality and loss less coding in the DCT coding is discussed for future use.

View full abstract

The algorithms standardized by ISO/IEC JTC1/SC29/WG11 for audio data compression (MPEG/Audio Coding Algorithms for short) are overviewed. The standard has a layered structure with Layer I, II and III algorithms. Layers I/II provide satisfactory coding quality with reasonable complexity, while Layer III dedicates additional complexity to the best possible quality. Two typical compression techniques for audio signals, the subband coding and the transform coding, are introduced. Key technologies inevitable for the coding algorithms, such as psychoacoustic weighting, block-size selection, MDCT, are covered one after another. Algorithms of Layers I through III including optional joint stereo coding are discussed with their subjective qualities. Fast algorithms for less computation and compensation for distortion caused by errors are also presented. A future ISO standardization schedule with important topics concludes this article.

View full abstract