Review on Text Recognition in Natural Scene Images

Text recognition is a field to detect text from digital images and produces meaningful information for various applications such as navigation, object recognition, hoardings, image retrieval. This paper presents a detailed review of various existing schem

  • PDF / 605,328 Bytes
  • 15 Pages / 439.37 x 666.142 pts Page_size
  • 101 Downloads / 200 Views

DOWNLOAD

REPORT


Abstract Text recognition is a field to detect text from digital images and produces meaningful information for various applications such as navigation, object recognition, hoardings, image retrieval. This paper presents a detailed review of various existing schemes performing the text recognition operation in natural scene images. Though researchers have done phenomenal work in this area but still there is scope to ample the prick. Keeping this in mind, a comparative and quantitative analysis of various existing schemes are presented. This review work is mainly focused on connected component and region-based methods. Various advantages and limitations are also discussed in this paper. The main aim of this paper is to present literature review for researchers to serve the purpose of good reference in the areas of scene text detection and recognition.



Keywords Text recognition Region-based and connected component-based methods Natural scene images Text detection Complex background







1 Introduction Text recognition serves as one of the important areas of pattern recognition and artificial intelligence. Text detection and recognition in natural images which are captured by camera is an important problem since text available in an image provides the useful information. Text present in images serves as an important source of information about the location from where they were clicked. Text recognition is the procedure to categorize the input character in accordance to the predefined character class. Text can in the form of scanned handwritten document or typed text

D. Kaushik (&)  V.S. Verma Ajay Kumar Garg Engineering College, Ghaziabad 201009, India e-mail: [email protected] V.S. Verma e-mail: [email protected] © Springer Nature Singapore Pte Ltd. 2018 B. Panda et al. (eds.), Innovations in Computational Intelligence, Studies in Computational Intelligence 713, https://doi.org/10.1007/978-981-10-4555-4_3

29

30

D. Kaushik and V.S. Verma

or a union of both. The text recognition system helps to establish a good communication between human and a computer. In the past few years, the focus on research work is increased to detect and recognize text from natural scene images in order to attain good truthfulness. It is basically possible as its utility in various applications, such as strengthening automobile driving, navigation, and recognize objects. Moreover, it strengthens the ability of image recovery models to recover important information as it yields explanation to the matter of images. Various approaches [1–5] have been developed so far for text detection, but still it holds to be as a interesting and unexplained issue. The reason is that the text present in scene images gets affected by difficult background, appearance dissimilarity, complex colors, different font size used in same word, uncommon font styles, inconsistent skew, various effects, and occlusion as illustrated in Fig. 1 which shows a broad range of images comprising different texts with variations discussed. Hence, we can see a lar

Data Loading...