jaidedai/easyocr

EasyOCR is a Python library that reads text out of images. OCR stands for Optical Character Recognition, the process of turning a picture of words into actual editable text. The problem it solves is that extracting text from photos, scanned documents, or screenshots normally requires complex setup; EasyOCR aims to make that as simple as a few lines of code. It supports over 80 languages, including Latin scripts, Chinese, Arabic, Devanagari, and Cyrillic. You point it at an image file (or pass in a URL, or a raw image object), tell it which languages to detect, and it returns the recognized text along with the position of each piece of text in the image and a confidence score. Internally it combines two deep learning models: a detection model that finds where text is in the image, and a recognition model, a CRNN, which stands for Convolutional Recurrent Neural Network, that reads the text at each location. The recognition model chains together a feature-extraction step, a sequence-labeling step using LSTM (a type of recurrent network good at ordered data), and a decoding step. All of this runs on PyTorch. Someone would use EasyOCR when they need to extract text from real-world photos, receipts, street signs, screenshots, or any image containing writing. It works on CPU if no GPU is available, making it accessible on ordinary laptops. Installation is a single pip command.