{ "cells": [ { "cell_type": "markdown", "metadata": {}, "source": [ "# 1 Introduction and Overview of PP-OCR System\n", "\n", "The first two chapters mainly introduce the DBNet text detection algorithm and CRNN text recognition algorithm. However, for an image in our actual scene, if we want to be based on the text detection or recognition model alone, it is impossible to obtain the text position and text content at the same time. Therefore, we connect the text detection algorithm and the text recognition algorithm in series to construct the PP- OCR text detection and recognition system. In the actual use process, the detected text direction may not be the direction we expected, which will eventually lead to text recognition errors. Therefore, we have also introduced a direction classifier in the PP-OCR system.\n", "\n", "This chapter mainly introduces the PP-OCR text detection and recognition system and the optimization strategies involved in the system. Through the study of this lesson, you can obtain:\n", "\n", "* PaddleOCR strategy tuning skills\n", "* Optimization techniques and optimization methods for text detection, recognition, and direction classifier models\n", "\n", "The PP-OCR system has undergone a total of 2 optimizations. The following is a brief introduction to the PP-OCR system and these 2 optimizations.\n", "\n", "## 1.1 Introduction to PP-OCR System and Optimization Strategy\n", "\n", "In PP-OCR, if you want to extract text information from an image, you need to complete the following steps:\n", "\n", "* Use the text detection method to obtain the polygon information of the text area (the text detection in PP-OCR uses DBNet, so the four-point information is obtained).\n", "* Perform cropping and perspective transformation correction on the above-mentioned text polygon area, convert the text area into a rectangular box, and then use the direction classifier to correct the direction.\n", "* Perform text recognition based on the rectangular box containing the text area, and get the final recognition result.\n", "\n", "The above has completed the text detection and recognition process for an image.\n", "\n", "The system block diagram of PP-OCR is shown below.\n", "\n", "
\n", " \n", "
\n", "
\n", " \n", "
\n", "
\n", " \n", "
\n", "