CONVOLUTIONAL NEURAL NETWORK
I .DISADVANTAGES OF USING ANN FOR IMAGE CLASSIFICATION
1.Too much computation - To handle variety in digits we can use simple artificial neural
network (ANN). But when you have bigger image, Eg: the image size is 1920 x 1080 x 3 ,
this nearly contains huge number of neurons and weights which is difficult to compute .
2.Treats local pixels same as pixels far apart – if you have a face of an animal in the left
corner versus right corner, it is still that animal’s face. Doesn’t matter where the face is
located.
3.Sensitive to location of an object in an image - If the pixels are moved around, it should
still able to detect the object in an image but with ANN its hard.
II. HOW HUMANS RECOGNIZE IMAGES
When we look at koala’s image we look at the features like its round eyes , black nose , fluffy
ears and we detect these features one by one. In our Brain there are different set of neurons
working on this different feature recognition of an image. These neurons are connected to
another set of neurons which will aggregate the results. If the features are eyes, nose and ears
then it is the face. And if legs and hands are recognized then it is the body part. There are
different set of neurons connected to these neurons which will again aggregate the result
saying that , if the images has koala’s head and body it means it is koala’s image .
III. HOW CAN WE MAKE COMPUTERS RECOGNIZE THESE TINY FEATURES .
We use the concept of filters. We take our original image and will apply a convolution
operation or a filter operation which results in a feature map .The benefit here is wherever
you see a number ‘1’ or a number close to it , which results in detecting a feature of an
image . “ Filters are nothing but the feature detector “ .
Location are invariant ( it can detect eyes in any location of the image ). With filters we get
different feature maps which are stacked together and they almost form a 3D volume for head
and body of the animal separately. This 3D volume is then flattened them to convert it into
1D array . These 1D arrays are joint together to make a fully connected dense neural network
for classification .
IV. WHY DO WE NEED A FULLY CONNECTED DENSE NEURAL NETWORK
HERE?
Neural networks are used to handel the variety in the inputs such that they can classify those
variety of inputs in a generic way .Feature extraction and classification are done till now but
there are 2 more components
1.ReLU operation
I .DISADVANTAGES OF USING ANN FOR IMAGE CLASSIFICATION
1.Too much computation - To handle variety in digits we can use simple artificial neural
network (ANN). But when you have bigger image, Eg: the image size is 1920 x 1080 x 3 ,
this nearly contains huge number of neurons and weights which is difficult to compute .
2.Treats local pixels same as pixels far apart – if you have a face of an animal in the left
corner versus right corner, it is still that animal’s face. Doesn’t matter where the face is
located.
3.Sensitive to location of an object in an image - If the pixels are moved around, it should
still able to detect the object in an image but with ANN its hard.
II. HOW HUMANS RECOGNIZE IMAGES
When we look at koala’s image we look at the features like its round eyes , black nose , fluffy
ears and we detect these features one by one. In our Brain there are different set of neurons
working on this different feature recognition of an image. These neurons are connected to
another set of neurons which will aggregate the results. If the features are eyes, nose and ears
then it is the face. And if legs and hands are recognized then it is the body part. There are
different set of neurons connected to these neurons which will again aggregate the result
saying that , if the images has koala’s head and body it means it is koala’s image .
III. HOW CAN WE MAKE COMPUTERS RECOGNIZE THESE TINY FEATURES .
We use the concept of filters. We take our original image and will apply a convolution
operation or a filter operation which results in a feature map .The benefit here is wherever
you see a number ‘1’ or a number close to it , which results in detecting a feature of an
image . “ Filters are nothing but the feature detector “ .
Location are invariant ( it can detect eyes in any location of the image ). With filters we get
different feature maps which are stacked together and they almost form a 3D volume for head
and body of the animal separately. This 3D volume is then flattened them to convert it into
1D array . These 1D arrays are joint together to make a fully connected dense neural network
for classification .
IV. WHY DO WE NEED A FULLY CONNECTED DENSE NEURAL NETWORK
HERE?
Neural networks are used to handel the variety in the inputs such that they can classify those
variety of inputs in a generic way .Feature extraction and classification are done till now but
there are 2 more components
1.ReLU operation
