Automatic image annotation based on learning visual cues

Labeling the semantic content of images is a problem known as image annotation. Automatic image annotation is the process by which a computer system automatically assigns metadata, in the form of captions or keywords, to a digital image. Automatic image annotation often uses computer vision techniques to extract meaningful cues from the image content. It can be applied in image retrieval systems to organize and locate images of interest from a database. Most of the image database systems employ manual annotation, that is users enter some descriptive keywords when the images are loaded,registered or browsed. Although manual annotation of image content is considered a “best case” in terms of accuracy (meaningful keywords are selected by the human) it is a time consuming and intensive process. In addition, human annotation is subjective: human observers tend to disregard background elements favoring subjects related on their life like humans or animals. Automatic image annotation can be regarded as multi-class image classification problem with a very large number of classes. Typically, image analysis in the form of extracted feature vectors and the training annotation words, are used by machine learning to automatically apply annotations to new images. The objective of this thesis is the development of an architecture of classifiers for automatic image annotation. The proposed framework is based on the idea that regionlevel analysis can improve image scene analysis and content description. The approach based on regions should provide a more robust classification since it is a good compromise between local and global approaches. The building blocks of my thesis will be: 1. Representing the image content by first segmenting the image and then computing meaningful descriptions for each region. 2. Studying data-driven feature selection for reducing the description dimension, devising unsupervised or partially supervised learning strategies to organize the data in homogeneous clusters 3. Designing an architecture of classifiers able to automatically assign a set of tags to a given image.