AWS Textract Integration for Document Data Extraction

Accounting departments spend days manually copying data from invoices and acts. Our clients faced the same problem: up to 80% of time wasted copying numbers from PDFs into ERP. Our AWS Textract integration service provides document data extraction using AWS OCR, enabling form and table recognition via the Textract API with Boto3 for key-value extraction and Queries for custom fields. We implemented AWS Textract—an OCR service that extracts not just text, but ready-made structures: tables, form key-value pairs, and identity document data. One project cut document processing from 15 minutes to 3 seconds; with 500 documents per day, that's annual savings of over 1.5 million rubles (approx. $20,000). For 500 daily invoices, our clients save $1,200 per month, with integration costs starting at $2,000. Below is how we achieve such results.

According to AWS Documentation, Textract models are trained on millions of documents and show >99% accuracy for standard fields.

Why Regular OCR Fails with Forms and Tables

Classic OCR engines return a stream of words with coordinates. Figuring out where a table starts and where a signature is—that's a weeks-long task. Textract uses neural networks trained on millions of documents: it automatically detects table boundaries, key-value relationships in forms, and even recognizes handwriting. For complex cases, the Queries mode is available—ask in natural language: "What is the total amount due?" and get the value with a confidence level.

What Does the Specialized Analyze ID Model Offer?

The Analyze ID model extracts data from passports, driver's licenses, and other identity documents with 99%+ accuracy. A confidence score is returned for each field, allowing you to filter out doubtful results. For example, for the DOCUMENT_NUMBER field with 99.8% confidence, you can use it directly without verification; at 85%, send it for manual review. This reduces the error rate to 0.2%.

How Asynchronous Processing Handles Large PDFs

The synchronous API (analyze_document) accepts up to 10MB and one page—ideal for streaming. The asynchronous API (start_document_analysis) works with PDFs up to 500MB. We use StartDocumentAnalysis with S3 triggers: file lands in a bucket → job starts → result saved to DynamoDB. For speed, we employ parallel requests via Lambda—throughput scales linearly.

import boto3
import json

class AWSTextractExtractor:
    def __init__(self, region: str = 'us-east-1'):
        self.client = boto3.client('textract', region_name=region)

    def extract_from_file(self, image_path: str,
                           feature_types: list = None) -> dict:
        """Synchronous processing of local file (up to 10MB, 1 page)"""
        if feature_types is None:
            feature_types = ['TABLES', 'FORMS']

        with open(image_path, 'rb') as f:
            response = self.client.analyze_document(
                Document={'Bytes': f.read()},
                FeatureTypes=feature_types
            )

        return self._parse_response(response)

    def extract_from_s3(self, bucket: str, key: str) -> str:
        """Asynchronous processing from S3 (for large files and PDFs)"""
        response = self.client.start_document_text_detection(
            DocumentLocation={
                'S3Object': {'Bucket': bucket, 'Name': key}
            }
        )
        job_id = response['JobId']

        # Wait for completion
        import time
        while True:
            result = self.client.get_document_text_detection(JobId=job_id)
            if result['JobStatus'] in ['SUCCEEDED', 'FAILED']:
                break
            time.sleep(2)

        if result['JobStatus'] == 'FAILED':
            raise RuntimeError(f"Textract job failed: {result['StatusMessage']}")

        # Merge pages
        pages = [result]
        while 'NextToken' in result:
            result = self.client.get_document_text_detection(
                JobId=job_id, NextToken=result['NextToken']
            )
            pages.append(result)

        return self._extract_text_from_pages(pages)

    def _parse_response(self, response: dict) -> dict:
        blocks = {block['Id']: block for block in response['Blocks']}

        # Extract forms (KEY_VALUE_SET)
        forms = {}
        for block in response['Blocks']:
            if block['BlockType'] == 'KEY_VALUE_SET' and 'KEY' in block.get('EntityTypes', []):
                key_text = self._get_text(block, blocks)
                value_block = self._get_value_block(block, blocks)
                if value_block:
                    value_text = self._get_text(value_block, blocks)
                    forms[key_text] = value_text

        # Extract tables
        tables = self._extract_tables(response['Blocks'], blocks)

        # All text
        lines = [b['Text'] for b in response['Blocks']
                 if b['BlockType'] == 'LINE']

        return {
            'text': '\n'.join(lines),
            'forms': forms,
            'tables': tables
        }

Basic Integration via Boto3

def extract_id_document(self, image_path: str) -> dict:
    """Specialized extraction from identity documents"""
    with open(image_path, 'rb') as f:
        response = self.client.analyze_id(
            DocumentPages=[{'Bytes': f.read()}]
        )

    result = {}
    for doc in response['IdentityDocuments']:
        for field in doc['IdentityDocumentFields']:
            field_type = field['Type']['Text']
            field_value = field['ValueDetection']['Text']
            confidence = field['ValueDetection']['Confidence']
            result[field_type] = {
                'value': field_value,
                'confidence': confidence
            }

    return result

# Example result:
# {
#   'FIRST_NAME': {'value': 'John', 'confidence': 99.5},
#   'LAST_NAME': {'value': 'Doe', 'confidence': 99.2},
#   'DATE_OF_BIRTH': {'value': '01/15/1990', 'confidence': 98.7},
#   'DOCUMENT_NUMBER': {'value': 'A12345678', 'confidence': 99.8}
# }

Extracting Custom Fields with Textract Queries

response = self.client.analyze_document(
    Document={'Bytes': content},
    FeatureTypes=['QUERIES'],
    QueriesConfig={
        'Queries': [
            {'Text': 'How much is the total due?', 'Alias': 'total_due'},
            {'Text': 'What is the invoice number?', 'Alias': 'invoice_number'},
            {'Text': 'Who is the supplier?', 'Alias': 'vendor'}
        ]
    }
)

Queries work in natural language—no need to write regex for each template.

Textract vs Classic Tesseract

AWS Textract outperforms Tesseract by 15% in form accuracy and 3x in table extraction speed. Manual entry takes 15 minutes per document, while Textract does it in 3 seconds—a 300x improvement.

To evaluate your case and get precise savings calculations, contact us for a free analysis of your documents.

Turnkey Integration Process

1. Analysis—we examine typical documents, identify fields and relationships.
2. Design—build a serverless pipeline with S3, Lambda, DynamoDB.
3. Implementation—write integration code using Boto3 and support Queries.
4. Integration—connect to your ERP or CRM via REST API.
5. Testing—run 100+ documents, achieve accuracy >98%.

What's Included

• Full pipeline documentation (IAM, S3, Lambda, DynamoDB).
• IAM role and security policy management.
• Training your team on using Textract results.
• One month of support after launch.
• Guaranteed 98%+ accuracy on your data.

Timelines and Cost

Basic integration with text extraction—3-5 days. If forms, tables, and Queries are needed—2-3 weeks. We provide a precise estimate after analyzing 10-20 of your documents. We are AWS certified with 5+ years of experience in document workflows—we guarantee that Textract will achieve 98%+ accuracy on your data. ROI typically occurs within 3-6 months thanks to an 80% reduction in manual work, which at average data entry costs yields savings of over 1.5 million rubles per year.

Contact us for a free demo: we'll evaluate your case and propose the optimal solution. Get a consultation today.

How Distribution Shift Kills CV Model Metrics in Industry

On a production line, a camera is installed to control product quality. The model is trained on 10,000 labeled images—test accuracy mAP 0.84. Deployed to production, and in the first week it misses 30% of defects. Lighting on the line changes between shifts; distribution shift nullifies the metrics. This is a classic story with computer vision in industry, where pattern recognition fails without proper drift handling.

Our engineers, with experience from 60+ computer vision projects, know how to eliminate such scenarios. We guarantee stable model performance under real conditions.

Object Detection: YOLO, RT-DETR, and Everything in Between

YOLO is the standard for real-time detection. YOLOv8 and YOLOv11 from Ultralytics are the most used versions in production: simple API, active community, built-in validation, and export to ONNX/TensorRT. For tasks with high accuracy requirements and less critical latency, RT-DETR, a transformer-based architecture without NMS, gives better mAP on COCO at comparable speed to YOLOv8l.

A typical mistake when training a detector: dataset of 8000 images, 3 classes, fine-tune YOLOv8m—F1 0.73 on validation. Look at confusion matrix—one class is almost never detected. Cause: imbalance 1:23. Solution: oversampling rare class, focal loss for objectness, augmentations (Mosaic, MixUp disabled for rare class as they "blur" it). Transfer learning is mandatory: pretrained on COCO weights reduces data requirement by 10 times. Fine-tuning on 500–2000 domain images yields a working model in 1–2 days on a single GPU.

For edge deployment: export to ONNX → TensorRT engine. YOLOv8n in TensorRT FP16 on Jetson AGX Orin gives 150+ FPS at P99 latency < 8 ms—3 times faster than ONNX Runtime without TensorRT. On server A10G: 700+ FPS for YOLOv8n in TensorRT INT8.

How Does Fine-Tuning YOLO Help in Pattern Recognition?

Suppose you need to find micro-defects on a metal surface—a task with high resolution and class imbalance. We use YOLOv8m pretrained on COCO and fine-tune on 2000 proprietary images. Apply augmentations Mosaic, MixUp, random perspective. After 200 epochs, mAP 0.5 reaches 0.93. Key techniques:

• Focal loss for the objectness head—reduces contribution of easily classified examples.
• Class-balanced sampling—equalizes representation of rare classes.
• Test Time Augmentation (TTA)—increases recall by 5–7% through averaging over flips and scales.

Get a consultation on architecture selection for your task—contact us.

Segmentation: SAM, Mask R-CNN, and Instance Segmentation

SAM (Segment Anything Model) from Meta changed the approach to segmentation. SAM 2 works with video, supports object tracking across frames—for interactive object selection by point or bbox, it's the best out-of-the-box choice. For production instance segmentation without interactive prompting, Mask R-CNN or YOLOv8-seg are used. YOLOv8-seg trains like a regular detector with additional masks, convenient in the same pipelines. Semantic segmentation (each pixel is a class) uses SegFormer, DeepLabV3+. SegFormer-B5 provides a good balance of accuracy and speed for satellite imagery or medical segmentation.

Case study: cell segmentation on microscopic images. Dataset of 400 images with manual annotation. Training Mask R-CNN on ResNet-50 backbone gave IoU 0.61—poor. Problem: objects (cells) overlap; standard NMS kills overlapping predictions. Solution: switch to cellpose (specialized architecture for biomedical tasks) + soft-NMS. IoU increased to 0.79.

OCR: When Tesseract Fails

Tesseract is a starting point for simple tasks: printed text, good lighting, straight layout. As soon as there are handwritten elements, non-standard fonts, perspective distortions, or multi-column layouts, Tesseract degrades quickly.

PaddleOCR is a production-grade solution: text block detection + recognition + structural analysis. Works out of the box for 80+ languages, including Russian. Supports tables and complex document structures. TrOCR (Microsoft) is a transformer OCR with strong results on handwritten text. For Russian handwritten text, fine-tuning is needed: the base model is trained mostly on Latin script.

What to Do When Tesseract Cannot Handle Pattern Recognition on Documents?

For tasks like "extract data from invoices/contracts/passports," we use LayoutLMv3 or Donut—these models understand document layout, not just text. Integration via Hugging Face Transformers, fine-tuning on 200–500 annotated documents. Typical pipeline:

1. Preprocessing: deskew, denoising, binarization via OpenCV.
2. Text block detection: PaddleOCR detection or CRAFT.
3. Recognition: PaddleOCR recognition or TrOCR.
4. Post-processing: normalization, validation via regex or LLM for structured fields.

For documents with fixed structure, template matching + OCR by coordinates is often more reliable than an end-to-end solution.

Face Recognition: Identification and Verification

Face recognition = detection + alignment + embedding + matching. Each stage matters.

Detection: RetinaFace or InsightFace for accurate face localization and keypoints. MTCNN is older but reliable. Embedding: ArcFace (InsightFace) is state-of-the-art for face recognition embeddings. Models iresnet50/iresnet100 pretrained on MS1MV3 (5M identities). Embedding vector 512 float32, comparison by cosine similarity. Threshold tuning: decision threshold is a critical parameter. At threshold 0.6, typical FPR on LFW benchmark is 0.001, TPR is 0.985. In production, threshold must be calibrated to the real distribution: people in masks, with changed appearance, different lighting conditions. Liveness detection is mandatory: MiniFASNet—lightweight model on CPU; FaceX-Zoo contains several pretrained liveness detectors.

Video Analytics

Video is a sequence of frames plus a temporal dimension. A naive approach—detecting on every frame—is expensive.

Tracking: ByteTrack and BoT-SORT are the standard for multi-object tracking. They work on top of any detector, adding persistent IDs to objects across frames—enabling object counting, motion tracking, velocity.

Optimization: not every frame needs processing. For static scenes, detect every 5–10 frames, with tracking in between. For event detection (person entering a zone), background subtraction (OpenCV MOG2) serves as a lightweight pre-filter before neural detection. Action recognition: SlowFast, VideoMAE for action classification. Heavy models—for production use ONNX export + TensorRT or offline processing.

How to Measure Pattern Recognition Model Quality in Production?

Quality monitoring is key to MLOps. We track:

• Prediction confidence distribution.
• Share of low-confidence predictions (indicator of OOD data).
• Drift of input images via feature distribution (embeddings from backbone).

A drop in average confidence from 0.87 to 0.71 over a week is an early signal of distribution shift. NVIDIA Triton Inference Server recommends tracking these metrics via Prometheus. Our certified engineers set up monitoring and guarantee SLA for inference quality.

Deployment of CV Models

For online inference, we use Triton Inference Server (NVIDIA)—production standard for serving CV models. Supports TensorRT, ONNX, PyTorch, dynamic batching, multiple instances. REST and gRPC API. We guarantee stable operation under load.

Edge deployment: ONNX Runtime on ARM/x86 CPU. TensorFlow Lite for mobile devices. OpenVINO for Intel CPU/GPU/VPU—gives 2–3× speedup on Intel hardware compared to ONNX Runtime. After deployment, we hand over the model with documentation and train personnel.

What Is Included in the Work

Deadlines and Cost

A prototype detector on existing data takes 1–2 weeks. Production system with optimization for target hardware takes 4–8 weeks. Full cycle including data labeling (1000–5000 images) takes 2–4 months. Cost is calculated individually for each task. Typical savings from implementing a quality control system can be significant per production line.

We have been in the market for over 5 years and completed 60+ computer vision projects. We will evaluate your project end-to-end—request a consultation to get a quote and technical proposal.