摘要
从二维图像还原三维形状的头部三维模型重构的实验中发现,配准三幅图像中所有点的计算量太大,于是提出了基于小波区域分解的方法;小波变换能够多尺度地从粗到精将头部地分离成头形、眼、鼻、嘴、痣、纹理等不同层次的特征区域,按精度对特征区域重建,减小配准时的搜索范围;对每个区域,使用灰度配准法则,然后用BP网络计算点深度,重构特征;这样精简了配准点数目,将处理速度提高了3～4个数量级。实验表明,该方法适应性良好,精度能达到1mm内。
The comeback way is based on 3 photos makes the computation potentially burdensome if every pixel to be matched, so a wavelet approach is designed to reduce it. Wavelet transformation can disassemble the head model into some areas at several levels, for examples, head outlines, eyes, nose, mouths, spots, veins, and so on. Disassembling is the first step, and then matching the feature pixels of the regions according to the desired precision, and later computing the Z values of these pixels with a BP neutral network. This paper proposes the whole scheme to reconstruct 3D head model supported by wavelet, which reduce the amount of matching points, and take much less time, and more reliable. The wavelet way is efficient approximately more than 1000 times and the error is no more than 1mm.
出处
《计算机测量与控制》
CSCD
2008年第4期573-575,578,共4页
Computer Measurement &Control
基金
上海市科技攻关重大专项(02D211019)
关键词
小波
立体视觉
重构
wavelet
3D vision
comeback
