Optimized Scene Text Detector and Paddle Optical Character Recognizer Techniques to Extract Text from Images

The increase in easily affordable and readily available camera embedded smart phones has enhanced the importance of tasks pertaining to computer vision and leverages endless possibilities in the field of image analysis. One of the most challenging and interesting areas in this field is the extraction of text from natural scene images. These are images that anyone can take at any given time without any controlled conditions. The diversity in these images makes it a difficult task for a machine to extract the texts from the images. This work applies two different approaches to the same problem to try and select the suitable method. Efficient and Accurate Scene Text Recognizer (ASTR) and Paddle Optical Character Recognizer (POCR) are the two methods that are applied. The Efficient and Accurate Scene Text Detection (EAST) algorithm provides better bounding boxes in complex scenarios and thus makes it a better text detection algorithm and combining that with the Paddle Optical Character Recognizer (PPOCR) recognizer gives good results.