الفهرس 
Introduction and Basic ConceptsOnly 14 pages are availabe for public view
Abstract
The growing availability of pen-based and touch-based hardware has recently resulted in an accompanying growth in the need for intelligent sketch-based user interfaces. The latter aim to combine the expressive power of free-hand sketching with the processing power of computers to understand what the user sketches; sometimes even before the user completes sketching. Sketching itself is a natural way of expressing and sharing ideas, as it allows a succinct transfer of concepts depicted on paper (or screens) to other forms of representing these concepts. The various qualities of sketching have caught the attention of application designers, who are starting to explore the advantages of intelligent sketch-based interfaces. The increasing availability of Tablet PCs and other hardware that supports pen-based interaction has led to an increased interest in interactive applications that can interpret hand-drawn sketches.
This thesis presents two methods for sketch recognition systems that are based on grouping and recognizing parts of physical objects to interpret a given sketch of such objects. This work intends to focus on recognizing the geometrical and informational properties of physical objects, particularly artefacts that commonly appear in the industrial community. The proposed methods are achieved by exploring two methods: one for the recognition of sketched objects that can be used as sources of information, and another for learning to apply this recognition to sketches of these objects.
The suggested methods are comprehensively evaluated on augmented variants of the TU-Berlin sketch benchmark, the QuickDraw dataset, and a proposed sketch dataset (RSO) for sketch classification and retrieval tasks. The experimental outcomes reveal that the proposed method brings about a substantial improvement over the state-of-the-art methods for sketch classification and retrieval. This work focuses also on improving the recognition performance that satisfies high sketch recognition accuracy and is speeded-up enough to provide a real-time interaction between the user and the computer in order to flexibly allow the user to draw freely and naturally.
This thesis consists of five chapters:
Chapter one provides an introduction to the thesis. It contains concepts of artificial intelligence, machine learning, pattern recognition, computer vision, and human computer interaction. This chapter contains definitions for sketching, sketch recognition, classification of hand-drawn sketching, it contains also goals of sketching recognition. Finally, it includes contribution of this thesis, and organization of this thesis.
Chapter two presents various hand-drawn datasets. In addition, it presents a survey for various hand-drawn recognition systems. The survey is classified into five groups: the history of hand-drawn sketching hardware; human computer interaction with sketching; applications of sketch-based interfaces; recognition methods; and learning methods.
Chapter three is devoted to present and explain a method for recognizing sketched objects by learning a specific set of geometrical features and creating a classification for these objects. This chapter consists of four sections. The first one is an introduction to the method. The second section illustrates the three steps of the method introduced in the chapter: extracting the geometrical features from a sketched shape, learning the geometrical features, and finally recognizing the sketched object by building a knowledge base containing the recognized objects with their geometrical features. The third section illustrates the experiments for the proposed method. The final section presents the experiment results and a discussion of them.
Chapter four is developed to present a proposed method for learning geometric features of hand-drawn sketches and is based on two models. The first model depends on a DT for recognizing and classifying objects in their categories. This model also learns to recognize sketched objects. The second model depends on FNN for training and testing; the model learns to recognize the objects and classify them into the correct categories. The proposed method works by extracting geometric features of sketched objects and classifying the objects into their categories. The method boosts the benchmark of sketch classification. The extensive experiments on the hand-free sketch benchmark datasets, the RSO dataset, the TU-Berlin sketch dataset, and the Quick Draw dataset; show great potentials of the proposed method.
Chapter five concludes all techniques presented in the thesis and suggests some ideas for future work.