Swarm Intelligence Methods for Unsupervised Images Classification: Applications and Comparative Study

Introduction

Users look for images every day on the net hoping that they find what can satisfy them. As the user scans the generated list from the entered query, they will select some images while ignoring others. Since the user is looking for a certain type of image during a specific query, the selected images must be related in some manner.

As Content-based image retrieval technology matures and is adopted by more users, another source of information for image retrieval becomes available, especially if we know that the number of images on the internet is estimated more than 15 trillion for that managing, analyzing and storing this gigantic image database has become a crucial challenge in the environment of computer science.

Several techniques have come out as the image-mining that represents all methods and techniques destined for the automatic image processing available in electronic format. In order to extract the knowledge and structured this information that must be filtered, prepared and classified to be a great aid to the decision making. One of the tasks of image-mining is the image classification. For example, we are confronted with a lot of images, hundreds or even thousands and asking a human who has no external help to classify them into groups with two conditions:

• 1.

  Images containing the same human gesture must belong to the same group;

• 2.

  Dissimilar images must belong to the different cluster.

It is a slow procedure and a difficult task that represent the content of our work along the machine. In order to help search engines to better satisfy the needs of users such as typing a query into Google Image: “Mike Tyson kick right”. We seek only the images of Mike Tyson with kicker right action. The solution obtained is shown in Figure 1 that is a screenshot of the first 51 images retained by Google, but we observe clearly that only7 images are relevant (bordered by the red) which represent kid drawing on a map.

Figure 1.

The result of the query “Zinedine Zidane run the ball” returned by Google image

If Google works with the principle of human gesture image classification so instead of treating all images, it will only analyse the cluster that contains the images with the run gesture. The world of image classification can be split into two large portions:

• •

  The supervised classification which requires the presence of a supervisor and hold a lot of drawbacks;

• •

  The class name must be known in advance this technology is less used caused by the number of images presented on the web where classes are not always known in advance;

• •

  The choice of the learning base which can greatly influence the calibre of outcomes;

• •

  Requires more external resources;

• •

  Little consistent classification;

• •

  The unsupervised images classification (clustering) using to solves the limitations of the previous technique that need only the presence of data entry based on a distance measure to discover the relationship between images with two objectives: minimization of the intra-class inertia and maximization of the inter-class inertia.

However, the classical clustering methods around this area are facing with some difficulties:

• •

  The images, representation and indexation;

• •

  The image size;

• •

  The choice of parameters (similarity measure and threshold);

• •

  The running time;

• •

  The initialisation of the cluster number.