Custom OCR for Text Recognition from Images

A client needed to recognize handwritten medical prescriptions from photos – ready-made solutions achieved less than 60% accuracy. A typical situation: the OCR pipeline fails on tilted or overexposed images, and specific terms (drug names) get distorted. We are a team of AI engineers with 5+ years of experience in computer vision, having delivered over 50 text recognition projects – we built a custom OCR model that raised accuracy to 93%. Here's how modern OCR works and how we adapt it to business tasks.

OCR (Optical Character Recognition) extracts text from images. The modern pipeline consists of three stages: detection of text regions → rectification → character recognition. Each stage affects final accuracy, and a weak link anywhere degrades the result. We use PaddleOCR as the base framework in 80% of projects for Cyrillic – it offers the best speed-quality balance among open-source solutions. Clients save up to 40% of their document processing budget through automation.

Which OCR framework to choose for Cyrillic?

We've tried all popular open-source solutions. For Russian, each has its strengths:

• PaddleOCR (PP-OCRv4) – accuracy 92.8% on ICDAR2015, best Cyrillic support among open-source. Suitable for production: fast on CPU, easy to fine-tune.
• EasyOCR – simple API, but for Russian accuracy is 5-10% lower, speed on CPU is 2-3 times slower.
• TrOCR (Microsoft) – transformer-based, achieves CER 2.89% on printed text. However, requires GPU and fine-tuning for Cyrillic.
• Tesseract 5 – classic, customizable for any font, but without custom training it loses to PaddleOCR on complex documents.

According to the official benchmark, PaddleOCR achieves 92.8% accuracy on ICDAR2015 (PaddleOCR GitHub).

Why is image preprocessing important?

OCR quality directly depends on the input to the model. A mobile phone photo – low contrast, noise, tilt. We apply a chain of transforms:

def preprocess_for_ocr(image: np.ndarray) -> np.ndarray:
    # Deskewing
    gray = cv2.cvtColor(image, cv2.COLOR_BGR2GRAY)
    angle = detect_skew_angle(gray)
    if abs(angle) > 0.5:
        image = rotate_image(image, -angle)

    # Denoising
    denoised = cv2.fastNlMeansDenoisingColored(image, h=10)

    # Contrast enhancement (CLAHE)
    lab = cv2.cvtColor(denoised, cv2.COLOR_BGR2LAB)
    l, a, b = cv2.split(lab)
    clahe = cv2.createCLAHE(clipLimit=3.0, tileGridSize=(8, 8))
    l = clahe.apply(l)
    denoised = cv2.cvtColor(cv2.merge([l, a, b]), cv2.COLOR_LAB2BGR)

    return denoised

Even simple deskew improves accuracy by 3-5%. For old scans with yellow background, we use adaptive binarization – Otsu or Sauvola. Preprocessing is especially critical for handwritten text: it boosts recognition accuracy by 15-20%.

How we do it: a case study of medical prescription recognition

Let's detail a real project from our practice. The task: accept prescription photos from a mobile app, recognize drug name, dosage, and instructions. Problems: handwritten text, blurry images, stamp overlaps.

Solution:

1. Preprocessing: CLAHE + binarization + shadow removal via morphology.
2. Detection: fine-tuned PaddleOCR detection model on 2000 labeled prescriptions (bbox labels).
3. Recognition: PP-OCRv4 recognition model fine-tuned on 50,000 synthetic prescriptions (generated with different handwriting styles).
4. Postprocessing: a drug dictionary (10,000 names) + LanguageTool for OCR error correction + LLM for context correction (0/O confusion).

Result: accuracy on test set – 93% (Character Error Rate 0.07). Processing time per image – 1.5 seconds on CPU. For comparison, Tesseract 5 without fine-tuning would give about 40-50% on such data – our pipeline was twice as accurate.

Process of work

Any OCR project goes through 5 stages:

1. Analytics: assess data, typical defects, domain dictionary.
2. Design: choose framework, pipeline architecture (queues, caching).
3. Implementation: write code, fine-tune models, integrate with your system.
4. Testing: measure accuracy on validation set, A/B test on real data.
5. Deployment and support: package in Docker, REST API or gRPC, monitor metrics.

What's included

• Comprehensive pipeline documentation describing all components.
• Trained model (weights + model card).
• Source code with launch instructions.
• Integration with your storage (S3, MinIO) and queues (RabbitMQ, Kafka).
• Training your team on the system.
• Accuracy guarantee (we fix metrics in the contract).

Timelines

Cost is calculated individually after data analysis. Get a consultation – we'll evaluate your project in one day. Contact us to discuss details and estimated cost.

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.