3

I am currently analyzing a set of pictures, that I want to classify. Classification is done by a Artificial Neural Network in a supervised manner. I have a test set that assigns to a each picture its class.

What I want to do now is generate a lot of descriptors and then do a PCA on these and do a statistical analysis how much the descriptor is able to describe the class of the picture.

How can i generate descriptors for these pictures programatically? This could help me in future classification problems too. Let us assume I have enough computation power (100 core cluster) Are there libraries that incorporate a lot of descriptors for images?

enter image description here

4

1 回答 1

3

您基本上可以按照两种方法开始：

  • 基于特征，使用SIFT或GIST等方法，然后是所谓的词袋方法。vlfeat 站点包含一个很好的演示。
  • 部署深度学习算法，例如稀疏自动编码器来学习数据集的基本特征，然后可以进一步用于分类。
于 2013-07-01T12:34:35.800 回答