Arabic Characters Recognition In Natural Scenes Using Sparse Coding For Feature Representations
2015 13th International Conference on Document Analysis and Recognition (ICDAR)(2015)
摘要
Character classification is the most important step in automatizing the process of reading text in natural scenes. The detection and the recognition of characters, needed in the end-to-end system, depend mainly on the robustness of the character classifier. Furthermore, Arabic characters classification need specific framework to deal with problems of their complexity and their diversity.In our work, we choose to use local feature representations since local features represent an efficient tool for dealing with problems of variability of size and color of text and problems of camera-based images.Adapting the Bag of Feature (BoF) technique to represent local features was extremely used in recent years. However, the BoF method removes the spatial information of local descriptors, which restricts the descriptive power for image representation. To solve this problem, we use Spatial Pyramid Matching (SPM). For the feature representation step, we choose to use sparse coding of Sift features.In this paper, we present a robust classification framework for Arabic Scene Text Characters (STC). Our architecture is based on the use of an adopted BoF model using SPM method and sparse coding to represent features.To evaluate our system, we propose a database of Arabic characters, called ARASTEC, since there aren't any such previous databases. Experimental results show the efficiency of this framework for Arabic STC recognition.
更多查看译文
关键词
Arabic Scene Character Recognition,Feature Representations,Sparse Coding,Spatial Pyramid Matching
AI 理解论文
溯源树
样例
生成溯源树,研究论文发展脉络