Due to the evolution of computer systems and theproliferation of Internet, numerous informationresources have been constructed. The deluge ofinformation makes the need to integrate informations,which are distributed o...
详细信息
Due to the evolution of computer systems and theproliferation of Internet, numerous informationresources have been constructed. The deluge ofinformation makes the need to integrate informations,which are distributed on the Internet and handled inheterogeneous systems. Recently, most of the XML-based information integration systems use XML DTD(Document Type Definition) for describing integratedglobal schema. However, DTD has some limitations inmodeling local information resources. Although W3C's XML Schema is more flexible andpowerful than XML DTD in specifying integrated globalschema, it has more complex problems than using DTDsin resolving conflicts. In this paper, we classify theconflict problems in integrating information resourcesusing XML Schema, and propose conflict resolutionmechanism using XQuery. Furthermore, we design anauthoring toolkit for supporting user to define globalschema easily with the schema definitions of localinformation resources.
This paper describes an in-home digital network architecturethat supports both real-time and non-real-time communica-tion. The architecture deploys a distributed token mechanismto schedule communication streams and to...
详细信息
This paper describes an in-home digital network architecturethat supports both real-time and non-real-time communica-tion. The architecture deploys a distributed token mechanismto schedule communication streams and to offer guaranteedquality-of-service. Essentially, the token mechanism preventscollisions to occur in the network, thus making the networkdeterministic. The distributed token scheduler uses a preemp-the earliest deadline first strategy, which guarantees a possi-ble bandwidth utilization of 100 percent. To allow non-real-time communication however, only part of the available band-width is allocated by the scheduler to real-time traffic, typical-ly 80 percent. The paper describes protocols to counter tokenloss and token duplication. The network is simulated and thepaper shows some results from this simulation. Based on low-cost ethernet hardware, a prototype of the network is built andtested. Last, the paper describes future directions.
This paper proposes a new method for location and poseestimation of active vision using the Stereo Omni-directionalSystem (SOS), which can acquire both color and real timestereo images in all directions of a *** SOS,w...
详细信息
This paper proposes a new method for location and poseestimation of active vision using the Stereo Omni-directionalSystem (SOS), which can acquire both color and real timestereo images in all directions of a *** SOS,wepropose a robust method to estimate the location and poseof the sensor by matching the edge histograms of the omni-directional images obtained at present location and thatobtained before hand at reference points. The proposedmethod first estimates the location and relative horizontalrotation angle by matching the vertical edge histogramsof the omni-directional images. Next, for each horizon-tal direction, we generate horizontal edge histograms of theomni-directional images obtained by the SOS,using the per-spective images with a FOV of 90°centered at the concerneddirection. We then match the horizontal edge histograms ineach direction and use the matching results to estimate thepose of the *** results show the effective-ness and robustness of the proposed method.
Development plans of high resolution image sensorsfor satellites are increasing these days. The data transmissioncapacity of onboard equipment is reaching up to several times aslarge as the down link circuit capacity ...
详细信息
Development plans of high resolution image sensorsfor satellites are increasing these days. The data transmissioncapacity of onboard equipment is reaching up to several times aslarge as the down link circuit capacity provided by presentsatellites. High-speed data compression is required in order totransmit the image data to ground stations in real-time. IDCP(Image Data Compression Processor) is space-born d?.compression processor with automatic bit-rate control function,which was developed under NASDA (National SpaceDevelopment Agency of Japan) contract for Advanced LandObserving Satellite (ALOS). A new radiation hardened ?ne-chip image data compression LSI (Large Scale Integrated Circuit),which has adaptive quantization function, was developed in orderto comply with high-speed data compression requirement as wellas constant bit rate transmission requirement.
In this paper, we propose a novel method using anadaptive discriminant space to identify people's faces fromvarious directions. The adaptive discriminant space thatis constructed from the candidates' facial im...
详细信息
In this paper, we propose a novel method using anadaptive discriminant space to identify people's faces fromvarious directions. The adaptive discriminant space thatis constructed from the candidates' facial imagesoptimally distinguishes between faces among certaincandidates. A principle method of our research graduallylimits the classification of categories by using facedirection estimation and recognition. Both the facedirection estimation method and the face recognitionmethod are appearance-based methods that employ LinearDiscriminant Analysis (LDA) on the Four-DirectionalFeatures (FDF). First, our method chooses the candidatesusing a hierarchical combination of face directionestimation and recognition. Next, face direction isestimated exactly by each candidate's discriminant spaceof face direction estimation. Finally, our method creates anew discriminant space from the candidates' facial imagesselected from the results of the face direction estimationprocedure. Limiting variations of a face direction canstrengthen a face recognition discriminant space. Usingthe adaptive discriminant space, the candidates can beselected optimally. Experiments showed that our methodimproved our accuracy rate 1.2% to 98.8% achieved byhierarchical face recognition of 105,000 images from 150subjects facing 35 different directions.
暂无评论