Arithmetic codes and the application to image coding were studied. At first, the idea of coding Markov sources by switching multiple block codes according to the initial state with sequential extension of the sources is studied, which resulted to confirm the advantage of arithmetic codes over block codes. Then several features are studied to find the correspondence between arithmetic codes and block codes, including the method to let the decoder exactly know the last valid symbol from ones being decoded. Finally, several design parameters of arithmetic codes are discussed including the binary expression methods of multi-symbol sources.

View full abstract

A photograph on ID card or passport is essential for its owner to certify himself as correct. However, there are some attacks to substitute or tamper the ID-photograph for another one illegally and thus we try to apply a watermarking technique to this problem. The watermarking techniques are already developed to various areas on image media presented in digital format, but not successful for the photograph produced in hard copy. This paper presents experimental investigations on the problem for watermarking the ID photograph, and discusses the technical approaches to prevent it from substituting or tampering.

View full abstract

Towards an intelligent image sensor with visual image processing function, preliminary bioelectronic image sensing system based on photoelectric function of protein bacteriorhodopsin was fabricated. The image sensing system consists of two functional parts: image detection including bacteriorhodopsin and supplementary feature extraction. Image detection, in which bacteriorhodopsin molecules are used as motion extraction elements, extracts image data relating only to moving object in movie images, whether background image was a pattern or not. The detected image data was a mix of different types of features, i.e., edge, moving direction, and region of moving object. The supplementary feature extraction part separates and categorize feature information from the detected images accordingly. Considering a total efficiency for feature extraction of moving object, we conclud that utilization of photo-response protein for intelligent image sensing system will enable us a realization of high-efficient intelligent image sensor.

View full abstract

‘Standard Object Colour Spectra Database for Colour Reproduction Evaluation (SOCS)', which contains about 50,000 object color reflectances/transmittances, was published as a Japanese Industrial Standard Technical Report(JIS-TR) in 19981). To promote, widen and standardize usage of SOCS, we selected representative data sets, including both typical sets and difference sets. Typical set samples have average characteristics of whole data in an object group, and difference set samples have metameric characteristics to corresponding typical set samples, respectively. This paper describes concepts and purposes of the selection, as well as the selection algorithms. A total of 365 representative data samples(235 samples for typical sets and 130 samples for difference sets)were selected and evaluated whether they meet the purposes for the sets. The evaluation result showed that they are very useful as follows.(a)Color correction matrices can be determined using only typical sets so accurately as they are determined using whole data in the groups.(b)Easy evaluation of color reproduction quality for color sensors by a combinatorial use of typical and difference sets, when spectral sensitivity of color sensors is known.

View full abstract

In case that we separate RGB color still images into R-, G- and B- color planes, it is known that the color plane images have highly correlation each other. It is also known that the images have different properties in a locally called context. In this paper, we propose lossless compression method for RGB color still images which is realized by removing these redundancies. In the proposed method, we generate color difference component of the prediction errors obtained from predictive transformation on each plane. Then we separate the prediction errors and the color differences of the prediction errors according to their context, and apply optimal entropy coding for each context. From the simulation results, we confirmed that the compression of the proposed method improved by 14% and by 13% in comparison with that of the LOCO-I and that of the CALIC, respectively, and was also equivalent to that of the CREW. And we also confirmed that the processing time of the proposed method was faster than that of these conventional methods.

View full abstract

(5, 3) reversible filter adopted in JPEG 2000 is calculated by using reversible lifting-based filtering. In this paper, it is shown that reversible lifting-based filtering can be replaced with real number operation and compensation calculation. By applying above method to lossless 4 points Walsh Hadamard transform, this transform is also replaced with real number operation and compensation calculation in the same way. As a result of studying characteristic of the compensation calculation, optimized lossless Walsh Hadamard transform can be constructed. We propose new lossless 2D-DCT which use this transform. It is shown that error of the proposed transform is less than that of conventional method.

View full abstract

In this study, a method to detect position and orientation of an object is proposed, in which the 3D Hough Transform is generated by utilizing the Hough Transform that is originally utilized for recognition of a line and the matching is three-dimensionally made between the three-dimensional information obtained by a laser range finder or stereoscopic range finder and the solid information of object. Detection result of six degree of freedom including rolling angle is shown in simulation.

View full abstract

When using the page turning machine, the enough space to establish must be secured. So it is difficult for persons who have serious handicap to use it. We attempt to solve the problem by taking the image of the book on the page-turning machine with a CCD camera and displaying it on a liquid crystal display. However, the distortion of the page is contained in the image. This paper proposes a method of correcting a distortion based on the shape of the upper end part and the bottom tip part of the page of book on the machine. By the experimenting results, we confirmed that it is possible to do appropriate correction when containing distortion even like a twist.

View full abstract

We propose a method to extract information about camerawork from a template-video of a person having a specific move pattern which was shot with a camera and to shoot others who appear in front of the camera with the same camerawork automatically. It is possible to shoot video of which the screen composition is changed by reflecting the cameraman's intention by reproducing the camerawork of the template-video. Our system gets information about camerawork for adjusting a geometrical-gap of a target and a time-gap of camerawork. Then, our system can shoot others with same camerawork using this information to adjust gap in real-time. In the shooting experiment, our system was able to shoot a video with the same camerawork as the template-video at the success rate of 70%.

View full abstract

We propose a robust car detecting and tracking system for bad weather conditions. The target scene of the system is highway scenes taken from the camera mounted in a car. The system is consisted of three main processes: car candidate area detection, car candidate area varification, and car tracking process. In the car candidate area detection process, image enhancement method is judged automatically to be added or not before edge detection, which is shown to improve the candidate detection precision in bad weather conditions. In the car candidate area varification process, both the grey level and time sequence information, and in the car tracking process both the edge and grey-level information, are combined to improve the robustness of the system in bad weather conditions. The system runs at a rate of 15 frames/sec using a PC with the IMAP-VISION realtime image processing board installed in the PCI bus. Encouraging results have been obtained through experiments using highway database scenes.

View full abstract

An image classification method with Chaos Neural Network (CNN) is proposed. CNN allows to find more close to a globally optimum solution than Back Propagation Neural Network (BPNN) due to the fact that behavier of the solution from CNN is chaotic so that the solution may get out from local minima. In order to determine the parameters required for CNN, training samples are used. The required parameters for the simulation are calculated beforehand using the training samples. Experimental results with simulation and actual satellite imagery data of Landsat TM show that the proposed CNN based image classification performance are better than BPNN performance by 18.2-54.3% for the simulation data while by 13.4% for Landsat-5 TM data, respectively.

View full abstract

Currently, most image retrieval systems are based on low level features of color, texture and shape, not on the semantic descriptions that are common to humans, such as objects, people, and place. In order to narrow down the gap between the low level and semantic level, object-based content analysis, which segments the semantically meaningful object on images, is an essential step. This paper describes a novel image similarity measure approach for image comparison at object categories. It is not only suitable for images with single objects, but also for images containing multiple and partially occluded objects. In this approach, the contour of objects is extracted, and feature is obtained from contours. A machine learning categorization algorithm is used to predict the category of each of object-contour segments. The image is represented in a k-dimensional space, where k is the number of categories of objects in all the images. Each dimension represents information about one of the category. The similarity measure between two images is computed using Euclidean distance between images in the k-dimensional space. Experimental results show that this approach is effective, and is invariant to rotation, scaling, and translation of objects.

View full abstract

Recently, human verification methods which use physical characteristics (biometrics) capture public attention. Face verification is one method. Main stream face verification methods used 2-D information such as light and shade images and contour images. But it couldn't get satisfactory verification rate if much constraint wasn' t put on users in the case of the data acquisition because they were poor against changes in illumination surroundings and the correction of the face direction couldn' t be done. Therefore, we propose a method which measures the 3-D face shape by light-section method and corrects direction of the 3-D face shape information. Then it verifies by using feature quantity that normal vector on the surface of the face is expressed as a complex number. In this paper, detail of the proposed method and result of the verification experiment to discuss effectiveness of the method are shown.

View full abstract

For the recognition of the human face, we have proposed a hybrid system, which combines a optical parallel joint transform correlator using Multi-Level Zone Plate Array (MLZPA) as a Fourier transform lens with a personal computer. In this paper, we present the design procedure and the fabrication process for an improved version of a compact parallel correlator (named COPaC II: 20×24×43cm³, 6kg) by redesigning the optics and the housing to solve thermal and vibrational problems. According to the guideline of biometrics for identification, we conducted the performance evaluation of the COPaC through one-to-one correlation with 300 database of front facial images taken with same day. As a result, we obtained considerably low rate error that False Non-match Rate (FNMR) is less than 1% and False match rate (FMR) is less than 1%. As a practical application of identifications, we experiment PC-log-in for 28 days using ID-less facial images with 10 persons database. The recognition rate was obtained 92.8% and rejection rate was 96.4% after 3 times of trials. Moreover, as a preliminary experiment, we checked eight changes in facial expression such as wink, smile, close eyes, it is distrainable if arrangement of a face is a not greatly different grade. Successful recognition results using the COPaC has been demonstrated with a throughput of 7face/s and we prospect as an innovative security system.

View full abstract

In this paper, a method of identification using image of hand which holds a pen is proposed. By combining with the handwriting recognition, It is able to realize the identification with the high security. By the camera, the hand which writes characters is picturized. Then, the feature quantity is extracted from information of the pen and shape, after direction and size are corrected, the identification is done using the subspace method. In addition, the result of the experiment for done examinees of 20 persons is shown in this paper in order to verify the effectiveness of this method.

View full abstract

In this paper, we propose a method of multiple vision with only one camera calibrated by using perspective projection and identification of vision lines. Several reference points are placed surrounding an object for the camera calibration. The camera parameters on different shooting position are calculated by minimizing the error between the actual coordinates of the reference points projected on image planes and the estimated ones. The multiple vision is given by unifying a number of stereo visions.

View full abstract

In this paper, we propose a new approach to building a scalable virtual environment by combining virtual sub-environments. To manage different types of users interaction, datasets, rendering methods, and so forth, we build and handle virtual sub-environments separately. To unify these sub-environments, we propose a use of an environment manager, which determines the users viewpoint of sub-environments according to the user's input, and compose display image from rendered images of them. By using the proposed methodology, we can combine sub-environments seamlessly, and build a virtual environment to manage different types of datasets, users interaction and rendering methods. By unifying sub-environments according to the user's viewpoint, we can construct a universal large-scale environment: the user can walk through cities, observe from above the earth, and fly through the solar system. Furthermore, we implemented the prototype system in CABIN and demonstrated the scalability of this architecture.

View full abstract

In recent years, the demand has been growing for multimedia archive systems that recognize the need for contents sharing and allow the secondary use of contents. This paper shows a method for constructing a future archive system that will allow a user to generate secondary contents by combining primary contents as desired through time-based editing. This method will also simplify contents management on the system side and provide for expansion of scale. The following new techniques supporting the secondary use of contents are reported here: (1) a composite-drive method that enables high-speed access through the dynamic and static use of optical media in a RAID configuration; (2) a storage-arrangement method for the initial parts of video segments that increases the simultaneous use of the archive system and the number of stored video segments; (3) a method of arranging media units (multiple media in a RAID configuration) on the basis of housing slots; and (4) methods of creating, storing, and searching for secondary contents through time-based editing centered around primary contents.

View full abstract

We have created a new advanced prototype system that creates an online “e-Learning on demand” environment with unique time-free, space-free and remote distance multimedia features. We propose a format and learning materials’ structure for constructing a one-source, multi-use material distribution system based on semantic transcoding technology. This technology enables us to provide a seamless material distribution process suitable to various learning environments. We confirmed the practicality of the electronic learning materials for lectures in a mobile environment as well as the effectiveness of one-source, multi-use electronic learning materials through various experiments which include wired note PC, wireless PDA and mobile phone.

View full abstract

We study an approach that applied the digital watermarking for the MPEG-4 bitstream error detection enhancement in video coding. There are some conventional error detection methods, such as the syntax violation detection in the decoding process and the method based on the DCT (Discrete Cosine Transform) coefficient features embedding in the DCT block. However they are not a conclusive one yet. This paper presents a new scheme which is much more practical than the conventional one. In the encoder, some features are extracted from the macroblock headers, motion vectors, and DCT coefficients, which are embedded in the quantized DCT coefficients as the digital watermarks. The decoder performs the accurate error detection through the watermark extraction. Simulation results demonstrate that the error detection ability of the proposed approach can be significantly improved, and that digital watermark embedding incurs little degradation to the picture quality.

View full abstract

This paper proposes a method to generate a good contrast output image from plural low contrast gray-scaled input images that are captured continuously by a fixed camera. The genetic algorithm is applied to allocate all gradations included in the accumulated input image to appropriate gradations in the output image nonlinearly. The positions to separate the range of gradations in the accumulated input image are coded into an individual chromosome. The individual fitness is evaluated by the summation of edge strengths that is one of spatial features included in the output image. The experimental results show that the good contrast images were generated by the proposed method in comparison with the linear gradations transform method and the histogram equalization method.

View full abstract

Normalized correlation matching is a typical method for template matching. However, the target image area to be searched must have the right position and the same scale as the template image. It may be required a template matching method to search the target image area that has free location, inclination and scale in an objective image. But, it needs large computational cost because two-dimensional pose parameters must be set in the sequential matching. Besides, a template image registered by the manual operation is not always optimal for the reliability and the computational cost for the image matching. This paper proposes a method to search the target image area efficiently that has free location, inclination and scale in objective image. The genetic algorithm is applied to both the template registration process and the image matching process to reduce the computational cost with maintaining the reliability. The experimental results show that the high-speed image matching was realized by the proposal method.

View full abstract

[in Japanese]

View full abstract