Multimedia content transcoding technologies have become more important than before for efficient content delivery in a wide variety of network and storage environment. In this paper, we discuss recent fast and high quality transocding techniques including coding bit rate conversion, spatial and temporal resolution conversion and coding format conversion for MPEG coded contents. As for the coding bit rate conversion, we show that a fast conversion can be realized through re-quantization process. When resolution conversion and/or coding format conversion are involved, we discuss a variety of studies to realize efficient transcoding by avoiding a computationally complex tasks such as motion estimation through the active use of coding information of input coded data.

View full abstract

This paper describes a new video content skimming technique based on automatic extraction of emphasized speech and an evaluation on which the technique is applied to video mail contents. The technique analyzes speech information in original contents statistically and estimates emphasized portions of speech track, by using prosodic parameters such as pitch, power and speaking rate. To extract emphasized portions of speech, the method introduces a metric, “degree of emphasis" that indicates how much emphasized each speech segment is. When a user requests the video content of the article, the method collects the video segments corresponding emphasized speech potion referring to the user specified ' digest rate' and creates the new video sequence. In the evaluation, we used two different kinds of video contents to apply the proposed technique; one is non-edited video such as home video and video mail, the other is fully-edited video such as sports and dramas. The result of the experiments show that the sequence of segmented video have enough quality in general; however, the technique cannot be applied sufficiently to fully-edited video contents.

View full abstract

In this paper, we propose “Orthographic Projective Grid Space (OPGS)" defined by two orthographic uncalibrated cameras. We also describe the methods to reconstruct the object in the space and generate the arbitrary view images without camera calibration. The coordinate system of OPGS is more similar to the one of Euclid space than the coordinate system of Projective Grid Space, in which the shape of the objects can be reconstructed without camera calibration. We use Silhouette Intersection method with octree data structure to make the reconstruction time faster. Additionally, we also describe the experiments and the evaluations about the space for demonstrating the effectiveness of the proposed OPGS. We can reconstruct the object as polygon model in the space and it looks like the models in Euclid space.

View full abstract

This paper describes a support application which enables the users to create the VR application in Spatial GUI System effectively. When the VR application designers create applications, it is often difficult to handle and understand many parameters and functions, both of which have to be set in the application program, with reference only to document manuals. This complication may interrupt the creative imagination and invite the inefficient coding of the users. In this paper, we describe our support application and evaluate the application for having some subjects complete VR programs, and we show that our application will provide great help for creating VR applications.

View full abstract

This paper describes a 3D modeling method for wide area outdoor environments, which is based on integrating omnidirectional range and color images. In the proposed method, outdoor scenes can be efficiently digitized by using two omnidirectional sensors; that is, the omnidirectional color and range images are acquired at the same position to register both of the images geometrically. In order to register multiple range images correctly, points on planar regions detected from the range data are used in registration process. Multiple range images are registered by making the corresponding planes overlap in different range images. The position and orientation of omnidirectional sensors acquired by a hybrid sensor consisting of RTK-GPS and gyroscope are used as initial values for registration, and rotations and translations of whole range data are optimized by registering range data simultaneously. The 3D model obtained by registration of range data is mapped by textures selected from omnidirectional color images in consideration of the resolution of texture and occlusions of the generated model. In experiments, we have carried out 3D modeling of our campus with the proposed method.

View full abstract

A method for enabling efficient facial data transmission over the communication network and reconstructing realistic 3D face images is proposed. A conventional method which combines the advantages of the model-based and image-based approaches to reconstruct 3D facial images with high efficiency and precision on a standalone environment is extensively used. In our approach, the face image recognition part and the 3D image reconstruction part, which were conventionally carried out on the same PC, are separated and implemented on two network-connected PCs. Also, by using JPEG compression, the size of image data transmitted over the network is reduced. Our method was implemented on network-connected PCs, and its effectiveness was confirmed. Subjective tests showed that image data could be reduced to around 6% of its original size.

View full abstract

Recently, the importance of SNR and spatial scalability in image coding has been increased. JPEG 2000 has advanced SNR and spatial scalability functions. However, the spatial scalability of JPEG 2000 is limited to the cases where decimated resolutions are 1/2ⁿ of the source image resolution. Therefore, Digital-Cinema resolution cannot be converted to SDTV or QVGA size image by decoding the JPEG 2000 bitstream directly. We propose a resolution conversion method using spatial scalability, which combines LL sub-band with a decimation filter. Experimental results confirm an improvement in PSNR of 1-2dB as compared to conventional resolution conversion methods.

View full abstract

In recent years, the opportunity to process digital image has increased because of the spread of mobile telephone and development of network. Above all, resolution conversion is one of the functions that are applied to digital images frequently. However, it has a problem that it causes distortion of digital image, because the image is expressed in discrete value.
In this paper, we propose a resolution conversion method that retains high quality. Our method regards a source image as a 3D shape by using Delaunay triangulation. As a result, our method reduces global distortion of resulted image after resolution conversion. In addition, it decreases block noise which is included in the image. Furthermore, our method reduces a processing cost to approximate the shape by using a 3D-Graphics board.
Finally, some experimental results show that our method is effective for the resolution conversion.

View full abstract

We propose a method for “digital scorebook for soccer game" which presented digitized soccer event and its visualization. The proposed method consists of estimation of player's position, event selection of play and its visualization. Our estimation of player's position allows for flexible camera movement and angle such as zooming in and out, pan, tilt and yaw. The reliable and robust estimation is based on image analysis by synthesis and Generalized Hough Transform. Also we classified soccer play events and annotate the game scenes using distance between players and ball. We demonstrate the applicability of these estimation and annotation scheme via visualization of animation character's behavior from the results.

View full abstract

Most of current ideas for making sports highlights are focused on scenes for getting a score and receiving loud cheer. However, depending on recent performances by Japanese players in other countries, Japanese TV programs often have opportunities of sending out their performance scenes as a highlight of the sports broadcasts. This paper presents a method for separating a sports broadcast content into units on the sports broadcast according to an analysis for telop regions by correlation coefficient. The proposed technique would be efficiently used as a preprocessing method in extracting the highlight from a sports broadcast content. The experimental results have shown that the proposed method has exactly separated baseball broadcasts into each inning and then each batting turn for every batter.

View full abstract

Now, these information searching methods are being studied using the characteristics of several media information, such as, text, audio, and video media. Then, we had proposed the video searching method for information identification or including data, when query video information is provided. As its characteristics, we use temporal histogram of luminance. But as the problems of our methods, there is the case when it can not solve the problem of search accuracy, because of the lack of moving information. Then, in this paper, we propose the video search method in local area in order to gain the search accuracy. At first, the procedure of our method is shown, and by the simulation experiments, the efficiency of our proposed method is indicated. Above the moments, by adapting the local area in our method, moreover, it is clear that the effect of speed up of searching process is appeared.

View full abstract

This paper describes an information service system which uses a signboard image captured with mobile phone as a key information. The image features of signboard have been stored together with image itself and related information which will be presented to users. Information will be retrieved by analyzing images sent via e-mail and comparing image features with ones stored in the database. A region of signboard is extracted by using gradient operation and Hough transformation. Image feature value so far used is a color histogram information in the signboard region. Experiments show that the correct information can be found in the 2-best retrieved results at the 90% probability when using images captured with a digital camera as keys. When using images captured with a camera of a mobile phone as keys, the performance is lower. However, this can be improved by calibrating the color characteristics of the mobile phone cameras.

View full abstract

We propose a new synchronization recovery scheme for watermark detection in moving pictures. Since most proposals that have so far tried to recover synchronization of still images could hardly solve such problems as computational complexity, additional distortions, or difficulty of application to images with affine transformation involving aspect ratio changes or cropping, synchronization scheme for still images cannot be simply applied to synchronization recovery for moving pictures. We have solved these problems on former schemes by using temporal features of moving pictures. In our method, the Temporal-Spatial spread spectrum technique is used for synchronization recovery, that is, temporally spread watermark signal is embedded into each frame. In a detection phase, temporal synchronization is recovered first and the embedded watermark signal appears as a geometrical pattern with a formal feature in an integrated image of inverse spread frames. The feature is used for recovery of spatial synchronization. We verified that our scheme worked well in an experimental environment modeling illegal copying that included HD-SD down conversion and recording by an analog VCR.

View full abstract

In this paper, the quality of recovered holographic images produced by CGH adopted JPEG2000 is investigated. As a result, it is observed that this process gives nice 3D reconstructed images. It appears to show that a transmission of 3D holographic images can be possible.

View full abstract

We propose one-directional traffic flow measurement method using double slit camera. Two slit cameras are installed in overhead location with longitudinal alignment. They shoot real traffic scene in downward direction. Slit camera outputs pseudo two-dimensional image that consists of space domain and time domain. We detect vehicles along time axis from information about pixel values at each line of a slit. Standard deviation is effective to detect bright color vehicles under the correct or over luminous exposure. We use change of standard deviation and average of a line of a slit as well as standard deviation to detect dark color vehicles under the under-luminous-exposure. We detect occupancy, time headway and time between two cars from each slit camera. In double slit configuration, we detect spot speed of vehicles by the time difference of its appearance at each slit. We estimate vehicle length by normalizing the vehicle region. We also divide vehicles into type of vehicle by length. We apply our proposed method into a real traffic scene. Experimental results show the effectiveness of our proposed method.

View full abstract

In this paper, we propose a new scheme of one dimensional traffic flow measurement using stereo slit camera. Both slit of stereo configuration are aligned perpendicular to the motion direction. Stereo matching of two line image gives a pseudo cross section of the moving object at one time. Compilation along time axis gives pseudo three dimensional or two and a half dimensional information of the object. This make separation task very simple, and is very useful passing vehicle counting. We applied the proposed scheme to passenger vehicle image and the result seems to be very promising.

View full abstract

A method for keyword extraction for retrievals of paper based documents and diagrams based on rule-based production system with certainty factor determination by means of Analytic Hierarchy Process: AHP method is proposed.
The method allows distinguish the documents into three categories, letter, form and drawing types of documents, then the most appropriate knowledge about keyword for retrievals, font size, location, frequency of the words etc. are selected for each document type. Production rules are created with more than five of the knowledge on keywords for retrievals. Traditional production system employs isolated knowledge so that it is not easy to take overall suitability of the knowledge. In order to overcome this situation, AHP is employed in the proposed system. Through experiments with 100 documents and diagrams, 98% success rate is achieved and also it is found that appropriate candidates for keywords with likelihood or certainty factor can be extracted with the proposed system. The proposed production system shows 50% of improvement on success rate of the keywords extraction from documents and diagrams compared to the existing production system without AHP.

View full abstract

In industrial image applications, a template matching method that can search a target image area in an objective image is often applied. This paper proposes a method to search a target image area with higher reliability even if a wide part of the image area is occluded. In the proposed method, a registered template image is separated into multiple block areas and an image pattern in each block is matched with an objective image. In the searching process, a weight that means a useful factor of a block area for searching is given to each block area. The total correlation rate between a template image and a partial image area in an objective image is calculated by the weighted correlation rates of all block areas. The proposed method was able to search all target image areas until 80% occlusion rate and recorded 100% recognition rate

View full abstract

Vector quantization (VQ) for image compression requires expensive time to find the closest codeword in both encoding and codebook design. In this paper, we propose a high-speed closest codeword search algorithm applicable to both encoding and codebook design for VQ including entropy-constrained vector quantization (ECVQ). By using a lighter modified distortion measure, we propose an appropriate topological structure of training vectors and codewords to eliminate unnecessary matching operations from the search procedure. This algorithm allows significant acceleration in the codebook design process. Experimental results are presented on image block data. These results confirm the effectiveness of our proposed algorithm.

View full abstract

The Remote color printing, where printing plate images are sent to remote printing factories by digital communication and printed there, necessitates the remote color proofing. High quality color proofs should be output at the printing sites. However, moiré phenomenon occurs, when the printing plate image is output using a proof printer whose printing resolution is different from its original printing resolution and resized halftone images are printed. This work analyzes the moiré phenomenon, and shows that the resizing does not cause the moiré phenomenon if the image has gray levels, but binarization causes halftone dot size shift and the dot size shift leads to the moiré phenomenon. Based on this understanding, a new algorithm (the gravitational re-allocation algorithm) to re-allocate halftone dots is proposed. In the algorithm, fractions of halftone dots are pulled against each other to form concentrated halftone dots, and no halftone dot parameters detection or image region classification are necessary.

View full abstract

In this paper, we propose a method of urban landscape reproduction, where landscape is rapidly reproduced far with a few polygons selected preferentially to maintain image quality. In this method, about each polygon of buildings, the importance of display is previously computed, based on visibility of polygons going with an observer's movement. In walkthrough, polygons that have high importance for moving direction are selected preferentially according to the computed results. By grouping buildings along with a road and selecting polygons of buildings that have high importance from each group, landscape can be reproduced far with less degradation of image quality, even in the case of selecting a few buildings. Through the evaluations of image quality and the effectiveness of data reduction based on the results of experiments on urban landscape reproduction and walkthrough using this method, we proved that our proposed method is practical.

View full abstract

Recently, image quality is improved by the research of sensation of reality that feeling of being at a live performance. Projection type holography especially is the techniques that make possible the natural 3D display. We are developing electro-holographic display system using water particle 3D screen. As a result, the viewing angle of reconstructed has been expanded by using water particle. But, it wasn't cleared that water particle screen shown flickers by gravity and flow. On the other hand, the technique with the flow controlled device is extremely effective to the improvement of flicker. So this time we propose the method for which the flow controlled device is used to the water particle screen. By the results of experiment we could improve flicker by this flow controlled device attached water particle 3D screen. And we could get high contrast 3D moving image.

View full abstract