This paper describes a system configulation and coding performance of a realtime HDTV CODEC which has been newly developed. In this CODEC, an HDTV frame is spatially divided into four frames which have a standard TV size respectively, and for each divided image is processed by an MPEG2 coding unit. This configulation enables not only HDTV coding but multiple standard TV coding. Moreover, a spatio-temporal signal shaping technique is introduced at lower bitrate transmission. The shaping process is adaptively performed based on coding situation. Through subjective evaluation test, it is clalified that high quality decodec images are obtained at 12〜18Mbps.

View full abstract

A motion estimation board for MPEG2 MP@ML encoding system has been developed. The board size is 220mm×233mm, and it contains at most four motion estimation LSI's MN85520(COMET) fabricated in Matsushita Electric Industrial co., ltd. The detection algorithm contains a telescopic search method, and both GDP structure and picture structure can be dynamically changed during a real-time encoding. Using this board, the high quality real-time encoding system for MPEG2 MP@ML can be realized.

View full abstract

This paper describes an efficient reversible coding scheme of digital map data for car navigation. The scheme separates the map into two parts that are roads and back grounds, and then it encodes them using two kinds of chain coding techniques suitable for their respective properties. This paper also proposes the following several methods to improve further coding efficiency. With regard to the road part. the number of pels to be encoded and that of starting points are reduced respectively by introducing new thinning processing and by adding new codewords which can represent crossings, forks and jumps. Moreover chain code sequence is encoded employing the ordering prediction and 2-D variable-length-codes. Simulation results indicate that these new methods decrease a coding rate by about 30%.

View full abstract

Although motion compensation schemes based on arbitrarily-shaped regions can reduce artifacts such as blocking-effects in a predicted image, these schemes generally require a large number of bits to encode region-shapes. To cope with this problem, this paper proposes a motion compensation scheme using Voronoi diagrams. Since a Voronoi diagram is reconstructed by only positions of points called generators, region-shapes can be encoded at a low coding rate. This paper also reports on some techniques for performance improvement of the proposed scheme, for example, introduction of the simulated annealing. Simulation results indicate that this scheme can produce better predicted images than the conventional block matching scheme.

View full abstract

This article describes past and resent activities of video coding in the International standardization bodies, ie, CCITT, CMTT and ISO/IEC. Based on the standardized coding systems, relationship between communication and broadcasting media and their transparency will become the most significant study items to be considered. In addition, according to the development of high speed and high efficient PC as well as WS, software based video coding technology will give shortly much more significant status compared with existing hardware coding.

View full abstract

This paper presents a new coding method for representing moving images with planes. We extend the layered representation and propose a new planar representation, called "setting representation, " to handle 3-D motion and 3-D depth information. Next, we introduce a consistent 3-D parameter extraction method which repeatedly applies conventional two frames extraction algorithm to multiple frames. The consistent 3-D structured information. however. cannot always be extracted under various conditions. In order to make good use of the extracted consistent 3-D information, we present a new video coding method based on 2-D/3-D structured information obtained by the integration of the layered and the setting representations.

View full abstract

We formerly reported on a method which extracted 3-D information of an object from the positional relationship between characteristic points corresponding to one another in image sequence by utilizing the perspective projection model. If the motion of the object includes rotational components, however, the method has difficulty in extracting accurately its 3-D information. To cope with this problem. in this paper approximate values of 3-D rotation parameters are derived from several successive frames on the basis of the property of vanishing points. Then by using these values as the initial ones of 3-D rotation parameters in the above method, the accuracy of extraction of 3-D information is improved, and moreover it also becomes possible to separate the motion of the object into the translational and rotational components.

View full abstract

For the flexible 3-D image communication, the describing method of 3-D images should be neutral for any kind of 3-D displays. For this purpose, a ray based method has been proposed and the information space of rays is named "Ray Space". In this paper, we propose an efficient projecting method of Ray Space in which rays are assumed to go straight on without any variation in the direction of propagation. We illustrate how the Ray Space is projected on a space of fewer dimension. and formulate three projecting methods. Our methods are useful for data compression of Ray Space and 3-D image data.

View full abstract

Reconstruction of virtual space has become very important in various applications. Computer graphics methods based on geometric models have been studied for this purpose. In order to obtain more natural representation, this paper proposes a method which represents and handles 3-D objects using a concept of "ray space." Handling of objects in the real space is equivalent to getting a sliced image with an appropriate plane in the ray space. The lack of vertical disparity in the ray space, however, causes the problem that the viertical magnification ratio cannot be determined. To solve this problem, we first segment multi-viewed images with disparity information. Then cluster centers of segmented regions are connected to form triangular patches. The vertical magnification ratio is calculated using the smoothed depth information obtained through triangular patches.

View full abstract