Dear Adnan,

The answer by Mr. Bandara is quite right.

What is SVM Training?

Think of how a child learns to identify objects. First we need to show him an object and tell them what it is. The same should be done many time until the child repeats the name of the object correctly when we show it to him.

SVM is kind a same, such that a raw brain which does not know anything about animal, cats or snakes. But we can show some samples of cats and snakes and let the SVM knows what we showed to it. This way the SVM learns to match the given input image to a label. This is called training. In another word, we input both the image and its label (the class of the image/target) to the SVM to get trained.

Do we actually input an image to an SVM?

The answer is NO (in general). How we human identify objects? normally to identify an apple which is on a rack of orange, we do not need to check on each an every atom or molecule of the fruits. First we check the shape (the oval shape). But both apple and orange are in oval shape. So the shape cannot discriminate between the two classes (apple and orange). Then we check the color, and now oranges are orange, but apples are red.

Similarly we do not need each and every pixel of the image. We need to obtain an abstract representation of the image like the shape, color and texture. This step is called feature extraction. In another word, we actually extract a selected set of feature and the feed it to the SVM. It means, SVM never see images but the abstracted version of images.

Why Abstraction Matters?

Some time the whole set of pixels may used as the input of the SVM. But most of the time we abstract the image. Abstraction or feature extraction benefits the process in many ways.

1. Size of the inputs will be reduced: 1000x1000 pixels can be reduced to a a few hundreds of values.

2. Invariant properties: The feature set may not be affected by the orientation, scale or illumination of the object.

3. Robustness: It allows to accurately classy many unseen instances of the objects. We cannot show all possible variations of an object to the SVM to get trained. But due to the abstraction step, most of the variations of the objects will be trimmed down to the same set of values in the feature set.

What are the Available Features?

There are many, and have to select one or many based on your requirement. Some examples are, color histograms, Histogram of Oriented Gradients (HOD), Speededup robust feature (SURF) and Scale Invariant Feature Transform (SIFT). First decide which feature (shape, color or texture) is contributed the most to discriminate the given set of image and the select the feature set based on that fact.

What SVM actually Does?

A feature is an N-dimensional vector and it is a point in N-dimensional space. Classification means we split the feature space by identifying walls which fairly cluster the set of features which belong to the same class together but apart from the feature set of other classes. SVM can identify these walls while in the training and when we give the feature of a new image, it checks with in which partition the given feature is mapped into and then return the label of that partition. I suggest you to go through the theoretical side of the SVM to know more about how it classifies the feature which are not linearly separable, by projecting them to higher dimensional space.

Dear Adnan,

The answer by Mr. Bandara is quite right.

SVM is just a classifier how it classify images.Is it necessary for me to covert image to any particular format?.please help .

Support Vector Machine (SVM) is a machine learning model for supervised data classification. SVMs essentially learn a hyper-plane which separates the data space into 2 regions (in 2 class case). In your case, suppose you have images of snakes and cats and you need to classify them. The steps you'll need to follow are

Extract 'features' from the images.

These 'features' may be functions of appearance of snake/cat in your case e.g colour of the animal, shape of the animal etc. By concatenating these features you can get a multi-dimensional feature vector.

Train an SVM classifier

Training essentially learns a separating hyper-plane between the feature vectors of snake class and cat class . For example, if your feature vector is 2-dimensional, training an SVM classifier would amount to 'learning' a line which best separates your labeled-data/training-data. You could use any of the multitude of freely available libraries of machine learning. In case you speak python, you could use sklearn for the task. This task of learning (hyper-plane in linear SVM) is referred to training.

Classify the images.

Once you have trained your model, you could then use it classify images whose class is not known

SVM is a classifier, It can be used to classify/distinguish between different animals.

For SVM to do our desired classification , we need to teach him. just like we teach kids to distinguish between different things. this teaching is called as 'training".

so what you have to do is to show several examples of different animals. suppose you want classify between cat and snake. you need to collect several images of different cats and snakes with true results (whether it is cat or snake). we call these true results as ''labels'.

Using this 'data' and 'labels', we can train SVM. and after training we can use it for classification of new images

thanks