Image Transformation based on Computer Vision to detect Multilingual Sign Board

Text written in Urdu is cursive and belongs to the same non-Latin family as Arabic, Chinese, and Hindi. Recognizing individual ligatures in natural scene photographs is difficult due to the difficulty of detecting and localizing Urdu text. The computer vision challenge of text recognition in photographs of natural scenes is difficult. Variations in text size, color, font, orientation, backdrop, illumination, and uneven lighting make text recognition in photos of natural scenes more difficult than optical character recognition (OCR). As a solution to the issue of identifying Urdu, Hindi, and English text in natural scenes, we propose using an Image Transformation technique in this piece of study. Text recognition in Urdu is more challenging than it is with non-cursive scripts due to a variety of factors, including different writing styles, many versions of the same letter, stretched and linked text, ligature overlapping, diagonal text, and condensed text. Implementing deep learning models for word recognition in natural scene photographs can benefit from the proposed image transformation techniques, which include rotation, translation, resizing, perspective transform, cropping, dilation/erosion, and Region of Interest (ROI) selection.