MRFScene: Multi-lingual Multi-oriented Scene Text Detection Using Markov Random Fields

The appearance of the text in images is of vital importance due to its capacity to communicate, effective accessibility of the information of visual content enhances comprehension. Text in images is helpful in identifying objects, understanding scenes and images for content-based image retrieval. Text identification is still a visionary process due to scene-specific issues such as appearance on curved surfaces, in different lighting situations, and varying font styles. The proposed approach uses the Gaussian blur technique to smoothen the uneven intensity values in the image. Based on the similarity nature, K-means method is used to arrange related types of pixels into appropriate groups. Markov Random Field (MRF) is used to establish the region-based correlation among textual and background portions in a picture. MRF assists in separating the main text pixels from the background region. The statistical property is performed to extract the actual textual candidates. The proposed procedure effectiveness is measured using precision, recall, and f-measure parameters.