CNN中的low-level feature 与high-level feature

• low-level feature:通常是指图像中的一些小的细节信息，例如边缘(edge)，角(corner), 颜色(color),像素(pixels),梯度(gradients)等，这些信息可以通过滤波器、SIFT或HOG获取；
• hight-level feature:是建立在low level feature之上的，可以用于图像中目标或物体形状的识别和检测，具有更丰富的语义信息。

通常卷积神经网络中都会使用这两种类型的features:卷积神经网络的前几层学习Low level feature, 后几层学习的是high level feature.

Quora上面也有这么一段解释：

Low-level features are minor details of the image, like lines or dots, that can be pickup by , say, a  convolutional filter (for reaaly low-level things) or SIFT or HOG (for more abstract things like edges). 

High levle features are built on top of low-level features to detect objects and shapes in the image.

CNN中的low-level feature 与high-level feature

原文：https://www.cnblogs.com/elitphil/p/12105825.html