赞
踩
Appearance和Shape
当然, 这是肯定和具体的任务具体的背景有关联的。
局部
在object detection中,broadly speaking, 先应用interest point detector, 然后应用feature descriptor to those point. 先检测感兴趣的点,再对感兴趣的点应用特征描述符。但是, 这是以前的老方法。This is what methods like SIFT, SURF, or more recently ORB and FREAK do.
全局
Global techniques instead try to describe the whole image or the whole object to be detected. One example are contour-based methods, which obviously are focused on describing and matching the object’s contour. 全局的方法不是找到几个点,也不是基于几个点,产生局部描述符, 而是利用整体轮廓.
根据Quora, I would say that appearance-based or texture-based are the same in this context. Shape features might be more related to describing the object’s contour. 所以,在他看来,appearance和texture基本一样, 而shape就是指的尺度上更大的轮廓。
Appearance: 颜色,
举例说明
From stackoom, 以足球为例, 形状是给定对象的几何形状,可以用矢量和边缘点来描述。 外观指的则是对象的纹理和颜色。(这就是为啥外观和纹理基本指的是一个东西。)
Copyright © 2003-2013 www.wpsshop.cn 版权所有,并保留所有权利。