Search

Video-on-demand (VoD) services allow users to receive real-time multimedia contents from afar distance. Yet, current architectures still require a context-aware indexing system to help users learn about story details. In this paper, we present a semantic processing system that offers structuralized video contents based on four basic entities: who, what, where, and when. Thus, users can select clips of interest among a video according to the guided clues. At the same time, we also present an SVC-compatible coding that improves spatial scalability in order to transmit the structuralized results for mobile devices efficiently. Users can, therefore, get a higher PSNR in low bit-rate applications. The experimental results have demonstrated the feasibility and effectiveness of our proposed system. Index Terms--Structuralized video contents, video mining, scalable video coding (SVC), video-on-demand (VoD).

This paper presents a speech interactivity embedded module (SIEM) that is quite simplified and suitable for programmable anthropomorphic dialogue and menu- driven recognition applications. The proposed SIEM adopts Average Magnitude Different Function (AMDF)-based pitch feature extractor and Dynamic Time Warping (DTW)-based recognizer. The simulated experiments indicate the performance is improved considerably relative to the conventional LPCC-based DTW method. In addition, we apply the proposed SIEM design for wireless master/slave architecture for mobile consumer devices, providing a cost effective solution to meet the request of real-time and flexibility in embedded system prototype development. It also encourages concurrent development of different parts of system hardware as well as module reusing~. Index Terms--embedded system design, speech interactivity, speech processing.

In this paper, a user friendly Mandarin speech dictation system based on neural network and language model is proposed. The proposed network is composed of dynamic programming model and Bayesian neural network which can treat the problem of time axis distortion and spectral pattern variation respectively. Besides, a statistical language model with hierarchical syntactical analysis is proposed to map the syllables into Chinese characters. The overall recognition rate of the system can achieve about 88.2%.

In this paper, a new hierarchical approach for recognition of the unconstrained handwritten numerals is proposed. In order to obtain a reliable skeleton of the observed character, some preprocessing operations including smoothing, noise removing, normalization, and thinning process are firstly applied to each character. Then, some interesting feature points are extracted from this reliable skeleton of the character. In the first stage of preclassification, a set of structural features named four-zone codes is adopted to preclassify the numerals. Due to the large degree of data and distortion of characters, a three layer of fuzzy neural network is used for fine classification. Experimental results show that a high recognition rate over 99.5% is obtained in this work.

The dynamic Huffman coding is a data compression technique, and famous for its flexibility. However, it has not been available for real-time applications because the adaption process is complex. In this paper, we proposed the architectures for dynamic Huffman coding. Both of the encoder and decoder are constructed by using the CAM-based memory modules. These memory modules parallelly perform the adaption process so that the coding can be efficiently executed. The encoder with the code table containing 32 symbols has been implemented. The simulation results show that the encoder can yield a input throughput of 40 Mbps operating at 40 MHz with 50% compression ratio.