Stil tanıma yoluyla zamandan kazanımlı çizim tanıma sistemleri geliştirme

Sketching is one of the natural mode of communication among humans. With the recent increase in the availability of pen-based devices, a growing trend towards sketch-based interfaces and sketch recognition systems have emerged in Human Computer Interaction. Modern approaches to sketch recognition make heavy use of machine learning technology to maximize recognition accuracies by learning from examples. Although having more training examples is key to the performance of any sketch recognition framework, certain aspects related to the practical use of machine learning technology have surfaced as real issues that need attention. One of these practical issues that hinders the development and deployment of sketch recognition systems is the excessive computational resources.During supervised learning of a sketch recognition system, if a large training dataset is used to train a system model, it costs more training time and results in a bulky model with poor classi cation performance. In this thesis, we propose a practical, simple, and easy to implement method that sketch recognition practitioners can resort to for partitioning their training data by based on sketching styles of users. Our method leverages the observation that certain groups of people have similar sketching styles, and generating models for smaller groups of people with similar styles reduces training and classi cation times without a signi - cant sacri ce in recognition accuracies.Our overall system is consisted of two main parts such that in the rst part, we partition the all available training data into style sub-groups and in the next part, we designed a system to identify sketching style of an incoming user to assign the user into one of the style groups generated in the rst part. We demonstrate the utility of our approach with empirical results obtained from databases of various sizes and characteristics.