The objective of stereo matching is to find the corresponding pixels from similar two or more images. However, it is difficult problem to get precise and consistent disparity under a variety of real world situations. In other words, the color values of stereo images are easily influenced by radiometric factors such as illumination direction, illumination color, and camera exposure. Therefore, conventional stereo matching methods can have low performances under radiometric conditions. In this paper, we propose a novel stereo matching approach that is robust in controlling various radiometric variations such as local and global radiometric variations. We designed a hybrid stereo matching approach using transition of pixel values and data fitting. Transition of pixel values is utilized for the coarse stereo matching stage, and polynomial curve fitting is used for the fine stereo matching stage. Experimental results show that the proposed method has better performances compared to the stereo matching algorithms of comparison group under severely different radiometric conditions between stereo images. Consequentially, we demonstrate that the proposed method is less sensitive to various radiometric variations, and shows an outstanding performance in computational complexity.

Reducing spatial redundancies between different camera views is one of the major challenges in multi-view video coding (MVC). However, with drastic increase in time consumption of such process, a proper way to reduce high computational complexity is needed. We propose an object-based fast prediction mode decision method in order to reduce complexity and time consumption of the coding process. Since a foreground object in fast motion tends to be predicted in disparity compensated prediction mode, we can properly decide which prediction mode to use for each part of the scene in advance. Segmentation process is carried out to identify the foreground object from the scene. The proposed method can save up to nearly half of time consumption in the encoding process without much loss in the quality of video sequence.

This paper proposes a proxy cache management policy based on patching scheme to service multimedia stream over the wide area network. With patching, only the initial portion of the requested object could be cached if there is only one stream for the requested object while the remaining part of the requested object is cached step by step as a function of time discrepancy among streams if more than two concurrent streams exist. By doing so, cache space allocation and replacement overhead could be reduced as compared to the case of object level space allocation and replacement; this eventually reduces the backbone network traffic to the origin server.

In this paper, we describe a highly available distributed video on demand (VoD) system, which is inherently fault tolerant. Multiple servers provide the VoD service. We propose techniques that detect and recover faults like those disk faults, server faults and system faults. And we present the server switching method that preserves reliability and real-time. When a server fails, it is replaced by another server in a transparent way; the clients are unware of change. The switching between the faulty server and the new server is smooth. The server switching method is alternate depending on the spare load rates of the surviving servers. Also we present the flow control mechanism so that the client keeps the proper buffer occupancy during such server-switching period. With fault tolerant capability, our VoD system can maintain a continuous service to client despite of faults.

Multicast transmission is one of the solutions to reduce the cost in large VOD systems. However, it is difficult to implement the interactive operations for an individual user in a multicast transmission system. We propose a frame based multicast transmission scheme to serve a large number of users and reduce additional channels while supporting VCR operations. This scheme supports both normal playback and VCR playback via the identical channels. The proposed FBMT is to deliver frames via appropriate channels based on frame type (I, P, B) of MPEG video. FBMT has superior performance over the multicast systems in terms of both initial blocking probabilities and VCR blocking probability. The initial blocking probability of the FBMT is decreased as the arrival rate is increased further, VCR blocking probability is always 0% to admitted users.

In the current Internet, the quality of service of audio application is not guaranteed due to variable loss and delay. Forward error correction (FEC)-based error control mechanisms which send the redundant information with the main information can be used to minimize the effects of packet loss. However, increasing the amount of redundancy may waste bandwidth and increase the end-to-end delay. The existing FEC-based error control mechanisms select the codec combination only based on the network loss rate. These mechanisms do not consider the end-to-end delay by increasing the amount of redundancy. This may result in exceeding the acceptable maximum delay. We propose a selection mechanism of the audio codec combination for error control. In simulations, we confirm that the proposed algorithm reduces the end-to-end delay and the packet loss together.

In this paper, we propose a proxy server management scheme to service continuous media objects on the wide area networks. In the proposed proxy management policy, base on the tree-like proxy server configuration, two physical proxy servers work as one logical cache. Moreover, each object is partitioned into two parts: the front-end and rear-end partition in order to reduce cache space management overhead that the cache space allocation algorithm works in two phases. In the first phase caching, two partitions of the requested object is cached in its corresponding physical proxy server of the logical cache: the front-end partition is cached in the child proxy server while the rear-end one in the parent proxy server. In the second phase, the rear-end partition of the requested object, which has been already cached in the parent proxy server in its first phase caching, will be cached again in the child proxy server of the logical cache depending on their access frequencies.

In this paper, we propose a PPC (Populari@-based Partial Caching) scheme for continuous media objects on the proxy servers. In this scheme, either the initial portion or the entire blocks of objects are cached and the size of the initial portion of each stream varies with the popularity of that stream. Under the proposed scheme, the initial latency otherwise happens to access the requested objects from the remote server can be reduced signijlcantly and moreover, the limited cache storage space can be utilized ef~cient~. Considering the characteristics of continuous media objects, we also apply the popularity of each stream by taking the amount of the data of eac~~ stream played back by clients into consideration. As a result of trace-driven simulations with 9 days of log jiles for VOD server, which were obtained from the website http://imbc.hananet. net, we show that, PPC, our proposed caching scheme, outperforms other well-known caching schemes in hit ratios and replacement overhead.

Continuous media objects, due to large volume and real time constraints in their delivery, are likely to consume much network bandwidth. Generally, proxy servers are used to hold the frequently requested objects so as to reduce the network traffic to the central server but most of them are designed for text and image data. The authors propose a two-layered network cache management policy for continuous media object delivery over the Internet. With the proposed cache management scheme, in each LAN, there exists one LAN cache and each LAN is further divided into a group of sub-LANs, each of which has its own sub-LAN cache. Each object is also partitioned into two parts: the front-end and rear-end partition. They can be loaded in the same cache or separately in different network caches according to their access frequencies. By doing so, cache replacement overhead could be reduced as compared to the case of the full size data allocation and replacement; this eventually reduces the backbone network traffic to the origin server.

We propose a frequency-based selective caching scheme where the buffer allocation unit varies with the access frequency of the requested object. With this scheme, buffer allocation works either at the distance level or object level. Once a request is admitted, it is sent to the disk(s) directly if the average arrival rate for the requested object does not exceed the marginal arrival rate. A distance between the requested stream and the preceding one accessing the same object is cached as a function of the length of the distance and the size of the buffer cache. An object is cached in its entirety, i.e., its caching mode can be changed into object mode if the average distance among the current streams for that object exceeds the average distance of the total number of current cached streams in the buffer cache. Our conclusion derived from the simulation is that our proposed caching scheme is more effective than other caching policies such as LRU and interval when it is applied to continuous media data.

The NOD article makes a difference to VOD data in terms of media type, size, creation interval and user interactivity. Because of these intrinsic characteristics, user access patterns of the NOD article can be different from that of VOD data. In this paper, we analyze the log file of one electronic newspaper to show the short-term popularity and long-term popularity patterns. Based on these patterns, we propose LLBF (Largest Life-cycle Based Frequency) prefetching scheme that uses the two popularity patterns to cache a set of popular articles. In Simulation, we show that the proposed LLBF prefetching scheme increases hit ratio, and reduces the number of replacements more than other replacement algorithms as a small number of articles such as headline news is prefetched in main memory.

This paper presents the design of a high performance MULVEC(M_U_L_tithreaded architecture for the V_E_ctor C_omputations), as a building block of massively parallel processing systems. MULVEC comes from the synthesis of the dataflow model and the extant superscalar RISC microprocessor. MULVEC reduces, using a vector wait queue and status field of each vector data, the number of synchronization, context switching, network traffic, and so on in case of repeated vector computations within the same thread segment. And if vector operand in one statement is more than three, MULVEC can be computed by non-strict method. After program having been simulated on the SPARC V9(super scalar bit RISC microprocessor), the performance (execution time of example program) of uniprocessor and MULVEC according to the different number of nodes are analyzed. The performance of MULVEC according to the different number of nodes are analyzed for the several programs.