An Impact of Frequency Domain Filtering Technique on Text Localization Method useful for Text Reading from Scene Images
M.J. Roopa, K. Mahantesh · 2019
As it is the era of digital life, we expect machines to read text, like humans from natural scene images. Even though plenty of works have been done on scene text extraction and recognition problem, it continues to exist due to varieties of challenges. These challenges influenced computer vision and image processing researchers to contribute novelty work for scene text reading (STR) applications. Finding out text location in an image is the supreme step of STR. State of the art algorithms were introduced during past few decades. Among successful algorithm, MSER is one which represents itself as more effective and efficient approach on ICDAR11 and ICDAR 2013 datasets. This paper mainly focuses on performance analysis of localization (MSER and SWT) algorithms on ICDAR 2015 Incidental text dataset before and after applying frequency domain sharpening filter. ICDAR 2015 dataset includes more complex images as it was taken without keeping text in mind. Image enhancement technique with MSER and SWT ensures high recall and Precision.