視訊資料瀏覽及檢索之研究

Abstract

在這篇論文中，我們發展一套簡單且實用的瀏覽視訊資料系統。此系統能 偵測出場景變化點，並將影片切割成幾個主要片段，且在每一片段中找出 一兩張之代表影片來。為了提升速度，DC-Image被使用在我們的系統。此 外，我們選擇了HLS 彩色模型並做了一些調整。我們也利用一些影像重要 特徵以作為日後影片檢索瀏覽之參考依據。之後我們也發展一套新的場景 變換偵測器，此偵測器可以同時偵測快速改變及漸進變化之畫面。 In this thesis, we present a simple but robust system for browsing video data.This system can automatically detect scene changes and divide a video stream into segments of "shots". Some representative images are then extracted from each shot and can be used as a helpful index for image browsing. In order to speed up the processing, the system operates on DC-images instead of fully decompressed images. Moreover, we analyzed several color models and found that the HLS color model is more suitable for our purpose. Besides, we also did some experiments and found that image features, like color percentage, color mass-center, and color covariance matrix, are very useful in describing the contents of images. Regarding scene change detection, we develop a new detector to detect both abrupt and gradual scene changes.