Master Double-Digit OCR with ddddocr: Deep Learning Library for PHP & Python

Introduction

ddddocr (Deep Double‑Digital Digits OCR) is an open‑source library that uses deep learning to recognize double‑digit numbers in images. It offers both training and inference capabilities, allowing developers to integrate accurate double‑digit OCR into their applications.

Background

In computer vision, recognizing numbers is common, but traditional methods struggle with two‑digit numbers because of visual similarity and overlapping features. ddddocr addresses this by combining convolutional neural networks (CNN) and recurrent neural networks (RNN) to achieve high accuracy on double‑digit tasks.

Features and Advantages

Deep Learning : Utilises CNN and RNN models for precise double‑digit recognition.

Open Source : Free to use, modify, and distribute, encouraging community contributions.

High Accuracy : Deep‑learning approach overcomes the limitations of traditional algorithms.

Flexibility : Provides both training and prediction APIs, allowing custom model training for various use cases.

Installation

pip install ddddocr

Basic OCR Usage

Single‑Line Text Recognition

Designed for images where text occupies the main area, such as alphanumeric captchas. Supports Chinese, English (case‑sensitive or forced case), digits, and some special characters.

Example file:

demo01.php

<?php
/**
 * @desc Basic OCR capability
 * @help https://github.com/sml2h3/ddddocr
 * @author Tinywan (ShaoBo Wan)
 */
declare(strict_types=1);

$ddd = PyCore::import('ddddocr');
$file = '../demo01.png';
$imgBytes = PyCore::bytes(file_get_contents($file));
$ocr = $ddd->DdddOcr(show_ad: false);
$res = $ocr->classification($imgBytes);
echo $res . PHP_EOL;

Input image:

demo01.png

Output:

# php demo01.php
2bghz

Another example image:

demo02.png

Output:

# php demo01.php
8A62N1

The library includes two OCR models; you can switch models by passing the beta:true parameter during initialization.

// Switch to the second OCR model
$ocr = $ddd->DdddOcr(beta:true);

When using the second model, the output may include additional promotional text and data source references.

Target Detection Capability

This feature quickly detects possible object bounding boxes in an image. It returns bbox coordinates (x1, y1, x2, y2) without performing OCR on the detected regions.

Example file:

demo03.php

<?php
/**
 * @desc Target detection capability
 * @help https://github.com/sml2h3/ddddocr
 * @author Tinywan (ShaoBo Wan)
 */
declare(strict_types=1);

$operator = PyCore::import("operator");
$builtins = PyCore::import("builtins");
$ddddocr = PyCore::import('ddddocr');
$cv2 = PyCore::import('cv2');

// Initialize with detection enabled, OCR disabled
$det = $ddddocr->DdddOcr(det:true, show_ad:false);
$f = PyCore::open('./demo04.png', "rb")->__enter__();
try { $image = $f->read(); } finally { $f->__exit__(); }

$bboxes = $det->detection($image);
PyCore::print($bboxes);
$im = $cv2->imread('./demo04.png');
foreach ($bboxes as $bbox) {
    [$x1, $y1, $x2, $y2] = $bbox;
    $im = $cv2->rectangle($im, [$x1, $y1], [$x2, $y2], color:[0,0,255], thickness:2);
}
$cv2->imwrite("./result.jpg", $im);

Input image:

demo04.png

Output (bounding boxes):

# php demo03.php
[[55, 344, 90, 378], [26, 345, 58, 379], [0, 345, 29, 377]]

Result image with drawn boxes:

result.jpg

If OCR functionality is not needed, initialize with ocr:false. To enable detection only, set det:true.

Slider Detection

The slider detection feature is implemented with OpenCV algorithms rather than AI. It can be disabled by setting ocr:false or det:false during initialization.

More information: https://github.com/sml2h3/ddddocr?tab=readme-ov-file

Related Resources

phpy – smooth integration of Python ecosystem into PHP

EasyOCR – simple and powerful OCR tool for PHP

ModelScope – open‑source large model platform

ModelScope training guide

Vision Transformer (ViT) implementation with phpy

phpy breakthrough for AI programming in PHP