Page 265 - Ai Book - 10

Activation map. With the help of feature map, we can do the following:

• Reducing the image size to make processing faster and efficient,
• Focus on several features that can help us in processing the image.
For example, biometrics or smartphones gather important information by recognising facial features like
eyes, nose and mouth instead of seeing the whole face.

u Rectified Linear Unit Function (ReLu): Rectified Linear Unit Function is the next layer of CNN after Convolution
Layer. As you know, feature map could be extracted from the Convolution layer which is then passed on to
the ReLU layer. The basic function of this layer is to remove all the negative numbers that exist in a feature
map. In other words, this layer introduces the concept of non-linearity function in a feature map.
ReLU is a non-linear activation function that can be commonly used in deep neural networks. This function
can be represented as:
f(x) = max(0,x)........................where, x= value if input
Here, you can see that the output of ReLu is maximum between the two values i.e., zero and input value. An
output is equal to zero when the input value is negative , else it is positive. Let us understand the concept of
negative and positive value with the help of an example. Suppose you have a 3×3 matrix of an input image:

–2 5 –2

0 3 –3
1 4 0

Here, you have seen that negative values exist in the matrix. These values can be removed by the ReLU layer.
After removing these values, output matrix will be:

0 5 0

0 3 0
1 4 0

u Pooling Layer: The working procedure of Pooling layer is similar to the Convolutional Layer. Basically, Pooling
layer is responsible for reducing the spatial size of the Convolved Feature while still retaining the important
features.
The Pooling layer plays an important role in CNNs because it can perform various kind of tasks like:
a. Makes the image smaller and manageable

b. Enhance the resistant power of image for small
transformations, distortions and transitions.
The two types of pooling which can be performed on an image
are as follows:

u Max Pooling : Max Pooling computes the maximum value or
the element from the portion of the image covered by the
Kernel. Thus, the output of max pooling layer is a feature map
that contains the most prominent features of the previous
feature map.

u Average Pooling: As its name implies, average pooling
computes the average of the elements present in the portion of the image covered by the kernel. Thus,
average pooling produces the average of features present in a patch.

139
139