2. Convolution Neural Networks

Convolution Neural Networks

Image Classification Pipeline

MLPs v/s CNNs

MLP	CNN
Only use fully-connected layers	Makes use of fully as well as sparsely connected layers
Only accepts vector as input	Also accepts matrices as input

MLPs

In a MLP, every hidden node needs to be connected to every pixel in input image. This accounts for a lot of redundancy.

CNNs

Instead of every node keeping information of every pixel of input image, we divide the image into 4 regions - red, green, blue and yellow. Then each hidden node can be connected to only the pixels in one of these 4 regions as depicted below.

Installation

To install opencv follow either of the below 2 methods -

1. Install OpenCV on Linux
2. Install OpenCV on Windows
3. If you have Anaconda Python installed, just run the below command -
   

conda install opencv

Note - You can validate installation of opencv by running, import opencv