A Study of Object Segmentation and Description for Motion Sequences

• Main Authors: Chao-Lung Lou, 駱昭隆
• Other Authors: Sheng-Jyh Wang
• Format: Others
• Language: zh-TW
• Published: 2002
• Online Access: http://ndltd.ncl.edu.tw/handle/83872791969472453216

碩士 === 國立交通大學 === 電子工程系 === 90 === In this thesis, we discuss two topics. One is about object segmentation in image sequences and the other is about shape description. In the first topic, we propose an improved motion segmentation algorithm based on Chien’s approach. In Chien’s approach, an image sequence is translated into the spatial-temporal domain and the motion segmentation issue becomes a 3-D object segmentation issue in the spatial-temporal domain. Following Chien’s approach, we further discuss the selection of contrast threshold and propose an adaptive scheme, to adjust the contrast threshold according to luminance and chroma conditions. In the second topic, we propose a new approach to track and describe segmented 3-D objects. In our approach, feature points of object’s shape are extracted first and the correspondence of these feature points are established frame by frame to describe the evolution of object’s shape along the temporal axis. Then, we describe the object’s spatial shape on each frame. After combining spatial description and temporal description, the shape of objects can be efficiently described.