Text Extraction from Book Cover Using MSER

Abstract

Detecting text from natural images is an ongoing field of research. In this paper, we propose a text-extraction and detection algorithm pipeline for obtaining information about a particular book by using computer vision. Features of the book such as its reviews, rating and, the price can be displayed to the end user, thus helping people make an informed decision about the book on which they are going to spend time reading. The text detection algorithm uses edge-enhanced Maximally Stable External Region for identifying the text-blob segments accompanied by various non-text area filtering algorithms to find the bounding boxes. These bounding boxes are then chained together and undergo OCR, performed by the Tesseract engine. The results of the extracted text are further improved by performing post-processing NLP techniques such as domain-based OCR and typo correction. The method proposed in this paper has extended use cases in different areas of text detection from natural images.