可調性視訊人臉追蹤

Abstract

如何從靜態的影像中偵測出人臉如何從影像中偵測出人臉的位置長久以來

一直是許多人所研究的問題，而動態的影片中來追蹤人臉更是更近一步的研究課

題。人臉偵測追常見的應用在於支援人臉辨識與人臉追蹤。目前人臉辨識多以單

張影像之辨識為主，很少有利用3D立體影像 進行人臉追蹤之研究。然而在人

臉追蹤的研究上，目前大部分的文獻都還只是靜態資料庫上的測試，都還無法有

效地拿來作即時性的應用，其中最大的困難就是不同的網路頻寬。

本論文提出基於可調式影像編碼下的人臉追蹤研究，在我們的系統架構下，整個

系統可以區分成兩個部分：第一個部分是針對可調式影像編碼和解碼的方法，第

二部份是對人臉追蹤的方法。我們希望以可調式影像編碼的方法來減少人臉追蹤

時的資料量，在資料量減少的情形下，不影響辨識的結果以及影像的正確性，因

為資料的減少便能大量縮減人臉辨識追蹤的時間，提高即時影像中人臉追蹤的能

力。為了減少處理資料量，我們使用移動向量(MV)來追蹤。

How is it measure people face to detect in static image, how be a lot of problem

that people study since detect the position that measures the faces of people for a long

time in the image and come to track in the dynamic film the faces of people are

research lesson of nearly one step even more question. To distinguish for main fact

pieces of image, few utilize 3D three-dimensional image carry on people research that

face track. But on the research that tracks in the faces of people, present most

documents are still only the test on static database, also unable to is it make instant

application of to bring effectively, most heavy difficulty different network frequently

wide among them.The paper is based on scalable video coding algorithm to do face

tracking on scalable video, first part we introduce the algorithm of scalable video,

second part is for face tracking method. We hope to reduce the data amount of

processing face tracking, we use motion vector (MV) to make it.