Project Description:

This project is part of the course on Data Mining and Machine Learning.

The topic of our project and research in this course would be Multimedia Mining. Multimedia Data essentially refers to data such as text, numeric, images, video, audio, graphical, temporal, relational and categorical data. Multimedia Mining is discovering knowledge from large amounts of different types of multimedia data. It involves the extraction of implicit knowledge, multimedia data relationships, or other patterns not explicitly stored in multimedia databases. We would like to work in this area as a part of our Minor Project and Major Project therefore we are still in a stage of defining a problem. A lot of work has been in done in the area of multimedia mining and a lot of scope still remains. Few topics of multimedia mining are as follows:

• Video Retrieval : Approaches are aimed at providing efficient browsing searching and retrieval of multimedia material. Application domains include large distributed digital libraries, broadcasting or production archives and video databases. The largest multimedia database is the WWW and specific approaches to this domain have been proposed by J.R. Smith and S.F. Chang. Applications include: video editing, video education and training, video database navigation.

• Image Classification for Content Based Indexing: Using Bayesian Classifiers, high level concepts are captured from low-level image features under the constraint that the test image does belong to one of the classes.

• Video Google : This is a Text retrieval approach to Object Matching in Videos. This work has been done be Andrew Zisserman and Josef Sivic. They have essentially described and approach to object and scene retrieval which searches for and localizes all the occurrences of user outlined objects in the video. An object is represented by a set of viewpoint invariant region descriptors using SIFT and then using a clustering approach to retrieve the object being searched for.

• Organising holiday images into meaningful categories : Andrew Zisserman and F.Schaffalitzky have worked in this area. Their work is: given an unordered set of images, divide the data into clusters of related image and determine the viewpoints of each image, thereby spatially or- ganizing the image set.

• Discovering objects in unlabelled images : The work here includes the discovery of object categories in a set of unlabelled images. Probabilistic Latent Semantic Analysis (pLSA) has been used here. Here the object categories and their approximate spatial layout are found without supervision given the images.

Our work right now is to get familiar with all the techniques available and identify a suitable set of problems which are open in Multimedia Mining.