Text Detection & OCR

Module: Computer Vision | Difficulty: Intermediate

Scene Text Detection

EAST (Efficient and Accurate Scene Text Detector)

Outputs per-pixel confidence and geometry:

Geometry encoding: for rotated rectangles.

CRAFT (Character Region Awareness)

Detects character regions and affinities between characters.

CRNN Text Recognition

CTC loss handles alignment between predicted sequence and target:

import torch
import torch.nn as nn

class CRNN(nn.Module):
    def __init__(self, num_chars=37):
        super().__init__()
        self.cnn = nn.Sequential(
            nn.Conv2d(1, 64, 3, 1, 1), nn.ReLU(True), nn.MaxPool2d(2, 2),
            nn.Conv2d(64, 128, 3, 1, 1), nn.ReLU(True), nn.MaxPool2d(2, 2),
            nn.Conv2d(128, 256, 3, 1, 1), nn.BatchNorm2d(256), nn.ReLU(True),
            nn.Conv2d(256, 256, 3, 1, 1), nn.ReLU(True), nn.MaxPool2d((2, 1), (2, 1)),
            nn.Conv2d(256, 512, 3, 1, 1), nn.BatchNorm2d(512), nn.ReLU(True),
            nn.Conv2d(512, 512, 3, 1, 1), nn.ReLU(True), nn.MaxPool2d((2, 1), (2, 1)),
            nn.Conv2d(512, 512, 2, 1, 0), nn.BatchNorm2d(512), nn.ReLU(True),
        )
        self.rnn = nn.LSTM(512, 256, bidirectional=True, batch_first=True)
        self.fc = nn.Linear(512, num_chars)
    
    def forward(self, x):
        conv = self.cnn(x)
        b, c, h, w = conv.size()
        conv = conv.squeeze(2).permute(0, 2, 1)
        rnn_out, _ = self.rnn(conv)
        return self.fc(rnn_out)

Key Takeaways

• EAST provides fast and accurate text detection
• CRNN + CTC handles variable-length text recognition
• End-to-end systems jointly detect and recognize text