Why we prefer Convolution Neural Networks (CNN) for Image data?

The answer of this Question hidden in the Architecture of the Convolution Neural Network which is quite uncommon than other neural neural networks.

the CNN have 3 main component's

1. Convolution layers
2. Pooling layers
3. Dense layers (Fully connected layers/ANN Architecture)

so because of the convolution layers and Pooling layers we prefer the CNN architecture for image data. But what exactly happens in those layers?

1) Convolution Layer :

This convolution layers operations help to extract local spatial patterns from the input image. By applying the filters across all over the image neural networks can detect the edges, texture, shape and complex pattens from the input image. The Hierarchical nature of the CNN allows them to learn complex features from the input images as the complexity of networks goes deeper.

(you can see in the image the filter of 2*2 is convolving with the image 5*5

that is dot product (element to element multiplication) of filter over the image pixels. output of this complete process is feature map which we pass through the activation function (commonly Relu) then this activated feature map we pass through the Pooling layer)

2) Pooling layers :

• Memory Efficiency In pooling layers we use most commonly max-pooling and average pooling helps to down-sample(reduce the size) of the feature map. This down-sampling reduces the spatial dimensions of the feature maps while retaining the most important information. which helps in memory efficiency and make architecture computationally more fast.

• Translation Invariance Pooling layers contribute to translation invariance by ensuring that the network focuses more on the presence of features rather than their precise location. This property is especially useful in tasks where the position of features may vary within an image.(The activated feature map pass through the pooling layer here we are using max-pooling. ) at last we get the output which we flatten (we can pass to again to convolution then pooling) then pass to the Dense layers and finally we gets output.