[RELEASE] iText pdfOCR 5.0.0

Author: iText-CI

.github/workflows/createpr.yml

@@ -0,0 +1,9 @@
+name: CreatePR
+
+on:
+  create:
+
+jobs:
+    call-createpr:
+      uses: XodoDocs/itext-github-workflows/.github/workflows/createpr.yml@master
+      secrets: inherit

.github/workflows/createrebase.yml

@@ -0,0 +1,14 @@
+name: "Create rebased branch"
+
+on:
+  pull_request:
+    branches:
+      - develop  
+    types:
+      - assigned
+      - synchronize
+
+jobs:
+    call-createrebase:
+      uses: XodoDocs/itext-github-workflows/.github/workflows/createrebase.yml@master
+      secrets: inherit

.github/workflows/licensecheck.yml

@@ -0,0 +1,16 @@
+name: LicenseCheck
+
+on:
+  push:
+    # 'branches-ignore' or 'branches' can be used to filter specific branches.
+    # By default, without any filters, it runs on every push to all branches.
+    # To be explicit, you can use:
+    branches-ignore:
+      - 'develop'
+      - 'master'
+      - 'rebased/*'
+  
+jobs:
+    call-licensecheck:
+      uses: XodoDocs/itext-github-workflows/.github/workflows/licensecheck.yml@master
+      secrets: inherit

easyOcr_to_onnx_export/README.md

@@ -0,0 +1,55 @@
+# easyOcr_to_onnx_export script
+
+### Disclaimer
+
+There is no official method for converting EasyOCR models to ONNX, so a custom approach was required.  
+The converted recognition models retain the same inputs and outputs as the original versions, while the detection models were slightly adjusted to better fit our use case.
+<br>
+
+
+## Setup Instructions
+
+Follow these steps to set up a virtual environment and install the required dependencies.
+
+### 1. Create a virtual environment
+
+```bash
+python -m venv .venv
+```
+
+---
+
+### 2. Activate the virtual environment
+
+* **Linux / macOS:**
+
+```bash
+source .venv/bin/activate
+```
+
+* **Windows:**
+
+```bash
+.venv\Scripts\activate
+```
+
+---
+
+### 3. Install dependencies
+
+Install all required packages using the `requirements.txt` file:
+
+```bash
+pip install -r requirements.txt
+```
+
+---
+
+### 4. Run the script
+
+```bash
+python easyOcr_to_onnx_export.py <model_dir>
+```
+
+Replace `<model_dir>` with the path to your EasyOCR model directory.
+

easyOcr_to_onnx_export/easyOcr_to_onnx_export.py

@@ -0,0 +1,113 @@
+#!/usr/bin/env python3
+import argparse
+import os.path
+
+import easyocr
+from easyocr import config
+from easyocr.craft import CRAFT
+from easyocr.detection import copyStateDict
+
+import torch
+
+
+detection_models = (
+    'craft',
+)
+recognition_models_gen1 = (
+    'arabic_g1',
+    'bengali_g1',
+    'cyrillic_g1',
+    'devanagari_g1',
+    'japanese_g1',
+    'korean_g1',
+    'latin_g1',
+    # FIXME: this one causes issues during export
+    # 'tamil_g1',
+    'thai_g1',
+    'zh_sim_g1',
+    'zh_tra_g1',
+)
+recognition_models_gen2 = (
+    'cyrillic_g2',
+    'english_g2',
+    'japanese_g2',
+    'kannada_g2',
+    'korean_g2',
+    'latin_g2',
+    'telugu_g2',
+    'zh_sim_g2',
+)
+recognition_models = recognition_models_gen1 + recognition_models_gen2
+
+
+# Detection model
+class TrimmedCRAFT(CRAFT):
+    def forward(self, x):
+        # Ignoring "feature"
+        y, _ = super().forward(x)
+        # Transposing result back to BCHW
+        return y.permute(0, 3, 1, 2)
+
+
+def get_detector(trained_model, device='cpu'):
+    net = TrimmedCRAFT()
+    net.load_state_dict(copyStateDict(torch.load(trained_model, map_location=device, weights_only=False)))
+    torch.quantization.quantize_dynamic(net, dtype=torch.qint8, inplace=True)
+    net.eval()
+    return net
+
+
+def main():
+    parser = argparse.ArgumentParser()
+    parser.add_argument('model_dir', help='directory with EasyOCR models')
+    model_dir = parser.parse_args().model_dir
+
+    for recognition_model in recognition_models:
+        print(f'Exporting {recognition_model}...')
+        gen = 'gen1' if recognition_model.endswith('_g1') else 'gen2'
+        filename: str = config.recognition_models[gen][recognition_model]['filename']
+        reader = easyocr.Reader(
+            lang_list=['en'],
+            gpu=False,
+            model_storage_directory=model_dir,
+            recog_network=recognition_model,
+            quantize=False,
+        )
+        # AdaptiveAvgPool2d cannot be exported to ONNX
+        # Specifying a static one instead assuming imgH=64
+        reader.recognizer.AdaptiveAvgPool = torch.nn.AvgPool2d((1, 3))
+        dummy_input = (
+            torch.randn(1, 1, 64, 512),
+            torch.randn(1, 512),
+        )
+        torch.onnx.export(
+            reader.recognizer,
+            dummy_input,
+            os.path.join(model_dir, filename.rsplit('.', 1)[0] + '.onnx'),
+            export_params=True,
+            input_names=('input', 'text',),
+            output_names=('preds',),
+            dynamic_axes={
+                "input": {0: 'batch_size', 3: 'width'},
+                "text": {0: 'batch_size', 1: 'batch_max_length'},
+            },
+        )
+
+    print('Exporting CRAFT...')
+    filename: str = config.detection_models['craft']['filename']
+    dummy_input = (torch.randn(1, 3, 2560, 2560),)
+    model = get_detector(os.path.join(model_dir, filename))
+    torch.onnx.export(
+        model,
+        dummy_input,
+        os.path.join(model_dir, filename.rsplit('.', 1)[0] + '.onnx'),
+        export_params=True,
+        input_names=('images',),
+        output_names=('y',),
+        dynamic_axes={
+            "images": {0: 'batch_size', 2: 'height', 3: 'width'},
+        },
+    )
+
+if __name__ == '__main__':
+    main()

easyOcr_to_onnx_export/requirements.txt

@@ -0,0 +1,3 @@
+easyocr==1.7.2
+torch==2.8.0
+onnxscript

pdfocr-api/pom.xml

@@ -5,7 +5,7 @@
   <parent>
     <groupId>com.itextpdf</groupId>
     <artifactId>pdfocr-root</artifactId>
-    <version>4.1.2</version>
+    <version>5.0.0</version>
   </parent>
 
   <properties>

pdfocr-api/src/main/java/com/itextpdf/pdfocr/AbstractPdfOcrEventHelper.java

@@ -1,6 +1,6 @@
 /*
     This file is part of the iText (R) project.
-    Copyright (c) 1998-2025 Apryse Group NV
+    Copyright (c) 1998-2026 Apryse Group NV
     Authors: Apryse Software.
 
     This program is offered under a commercial and under the AGPL license.

pdfocr-api/src/main/java/com/itextpdf/pdfocr/IImageRotationHandler.java

@@ -1,6 +1,6 @@
 /*
     This file is part of the iText (R) project.
-    Copyright (c) 1998-2025 Apryse Group NV
+    Copyright (c) 1998-2026 Apryse Group NV
     Authors: Apryse Software.
 
     This program is offered under a commercial and under the AGPL license.

pdfocr-api/src/main/java/com/itextpdf/pdfocr/IOcrEngine.java

@@ -1,6 +1,6 @@
 /*
     This file is part of the iText (R) project.
-    Copyright (c) 1998-2025 Apryse Group NV
+    Copyright (c) 1998-2026 Apryse Group NV
     Authors: Apryse Software.
 
     This program is offered under a commercial and under the AGPL license.
@@ -40,6 +40,7 @@ public interface IOcrEngine {
      * in the format described below.
      *
      * @param input input image {@link java.io.File}
+     *
      * @return {@link java.util.Map} where key is {@link java.lang.Integer}
      * representing the number of the page and value is
      * {@link java.util.List} of {@link TextInfo} elements where each
@@ -63,6 +64,35 @@ public interface IOcrEngine {
      */
     Map<Integer, List<TextInfo>> doImageOcr(File input, OcrProcessContext ocrProcessContext);
 
+    /**
+     * Reads data from the provided list of input image files and returns retrieved data
+     * in the format described below.
+     *
+     * @param inputs list of {@link java.io.File} input images
+     *
+     * @return {@link java.util.Map} where key is {@link java.lang.Integer}
+     * representing the number of the page and value is
+     * {@link java.util.List} of {@link TextInfo} elements where each
+     * {@link TextInfo} element contains a word or a line and its 4
+     * coordinates(bbox)
+     */
+    Map<Integer, List<TextInfo>> doImageOcr(List<File> inputs);
+
+    /**
+     * Reads data from the provided list of input image files and returns retrieved data
+     * in the format described below.
+     *
+     * @param inputs list of {@link java.io.File} input images
+     * @param ocrProcessContext ocr processing context
+     *
+     * @return {@link java.util.Map} where key is {@link java.lang.Integer}
+     * representing the number of the page and value is
+     * {@link java.util.List} of {@link TextInfo} elements where each
+     * {@link TextInfo} element contains a word or a line and its 4
+     * coordinates(bbox)
+     */
+    Map<Integer, List<TextInfo>> doImageOcr(List<File> inputs, OcrProcessContext ocrProcessContext);
+
     /**
      * Performs OCR using provided {@link IOcrEngine} for the given list of
      * input images and saves output to a text file using provided path.