Multi-View Object Recognition and Classification. Graph-BasedRepresentation of Visual Features and Structured Learning andPrediction.

Abstract:

Computer Vision is a subfield within artificial intelligence that includes methods for acquisition, processing, analysis and understanding of images to get results in numerical or symbolic form. The information provided by the results is used to make decisions.We do not speak ofComputer Vision in isolation, interaction with other fields is inevitable and deserve particular attention image processing, pattern recognition and Machine Learning. The main objective of this project is to analyze the behavior of visual feature extraction algorithms and their effectiveness in decision making. The detection of an object in an image, its classification and recognition are the type of decisions that are studied.

Feature extraction algorithms are applied to attempt multi-view object recognition. To tackle this problem a new approach is proposed. This approach creates a graph-based representation of the object using cluster analysis recursively. The nodes of the graph represent the main physical components that make up the object. Support Vector Machines (SVMs) are used to classify the nodes, thus classes are classified independently. Finally, the graph-based representation of the object is exploited to drop the assumption of independence and find relations between classes using Structured Output-Support Vector Machines (SO-SVMs).