Detection Character Regions and Comparison Features in Event flyer Images based on Machine Learning

Abstract

Tame-map is a smartphone application that allows users to easily share regional events information within their daily lives by uploading digital flyer images on the web. Although there are a lot of event flyer images uploaded daily, these data have been organized manually. This manual work is hard and takes a huge time, so a system to automate the manual work has been required. An automatic processing method will support the organization, such as categorization, assigning keywords to each image. To develop a basis of automatic flyer images processing, this paper examines detection character region in event flyer images based on machine learning. Concretely, we compared the two features which are obtained vectors based on unsupervised/supervised leanings, such as k-means and SVM respectively.