Document Authenticity Verification (Anti-Fraud Detection)

Fraudsters forge documents: digital (Photoshop, copy-paste), physical (laminate, ink), and live (covering part of the document, photo substitution). Banks, insurance companies, and government agencies face hundreds of such attempts daily. Our document authenticity verification system (Anti-Fraud Detection) uncovers forgeries using a multi-layered architecture combining classic metadata analysis, Error Level Analysis (ELA), convolutional neural networks, and font consistency checks. Traditional manual verification requires resources and time, and errors lead to losses — an average bank spends significant resources on manual verification; our system can cut those costs by 40–60% and achieve up to 96% accuracy on real data. We guarantee validation on your dataset.

Why Traditional Verification Methods Fall Short

Manual document checking is labor-intensive and prone to human error. An operator spends an average of 2–3 minutes per document, and under high workload, up to 15% of forgeries are missed. Rule-based automatic systems (regex, CRC checks) fail to detect complex manipulations: inserting a fragment from another document, altering digits while preserving the font, imitating security features. That is why an ensemble of machine vision and deep learning methods is needed — an anti-fraud system capable of detecting forgery at the manipulation level.

How Document Authenticity Verification Works

The system analyzes several aspects of a document: metadata, compression structure, visual elements, and fonts. Each stage provides its own signal, and the ensemble of methods boosts overall accuracy. Let's look at the key components.

Error Level Analysis (ELA)

This is a basic method: regions altered in JPEG will have a different compression error level. Python implementation:

import cv2
import numpy as np
from PIL import Image
import io

def error_level_analysis(image_path: str, quality: int = 95) -> np.ndarray:
    """ELA for detecting JPEG manipulations"""
    original = Image.open(image_path)

    # Re-compress with given quality
    buffer = io.BytesIO()
    original.save(buffer, format='JPEG', quality=quality)
    buffer.seek(0)
    recompressed = Image.open(buffer)

    # Difference
    ela_image = np.array(original, dtype=np.float32) - \
                np.array(recompressed, dtype=np.float32)

    # Amplify for visualization
    ela_image = np.abs(ela_image) * 10
    ela_image = np.clip(ela_image, 0, 255).astype(np.uint8)

    return ela_image

Why ELA Isn't Always Enough

ELA produces false positives on textures and high-quality photos. That's why we add a CNN-based detector. It is trained on real and forged documents and accounts for more features. The CNN detector is 1.5–2 times more accurate than ELA on complex textures.

from transformers import AutoModelForImageClassification

class ManipulationDetector:
    def __init__(self):
        # Model trained on real/forged documents
        self.model = AutoModelForImageClassification.from_pretrained(
            'path/to/manipulation_detector'
        )
        self.model.eval()

    def score(self, image_path: str) -> float:
        """Returns probability of manipulation [0, 1]"""
        # Input stack: RGB + ELA + compression errors
        rgb = load_and_preprocess(image_path)
        ela = error_level_analysis(image_path)
        features = np.stack([rgb, ela], axis=0)

        with torch.no_grad():
            output = self.model(tensor_from(features))
        return float(torch.sigmoid(output.logits[:, 1]))

Checking Security Features

Document security features: watermarks, holographic stickers, microprinting, guilloche (wavy patterns). For each document type, we describe expected visual features. We use FFT analysis to identify regular patterns. FFT analysis is 3 times more effective than visual manual inspection.

def check_watermark(image: np.ndarray, expected_region: dict) -> dict:
    """Check presence of watermark in expected region"""
    x1, y1, x2, y2 = expected_region.values()
    roi = image[y1:y2, x1:x2]

    # FFT analysis to detect regular patterns (guilloche)
    gray = cv2.cvtColor(roi, cv2.COLOR_BGR2GRAY)
    f_transform = np.fft.fft2(gray)
    f_shift = np.fft.fftshift(f_transform)
    magnitude = 20 * np.log(np.abs(f_shift) + 1)

    # Presence of characteristic frequencies in guilloche
    expected_freq_present = analyze_frequency_pattern(magnitude)

    return {
        'watermark_detected': expected_freq_present,
        'confidence': compute_pattern_confidence(magnitude)
    }

Font Consistency Analysis

Altering digits or letters in a document often gives away font inconsistency: different stroke thickness, different font size, different line spacing. We cluster character heights and identify outliers:

def check_font_consistency(ocr_words: list[dict]) -> dict:
    """Check font feature consistency"""
    # Cluster by character height
    heights = [word['height'] for word in ocr_words]

    # If a group of words has a significantly different height — suspicious
    mean_height = np.mean(heights)
    std_height = np.std(heights)
    outliers = [w for w in ocr_words
                if abs(w['height'] - mean_height) > 3 * std_height]

    return {
        'consistent': len(outliers) == 0,
        'suspicious_words': [w['text'] for w in outliers],
        'anomaly_score': len(outliers) / max(len(ocr_words), 1)
    }

Forgery Detection Accuracy

Accuracy depends on the type of fraud. Combining all methods, we achieve the following performance:

What Multi-Layer Verification Delivers

System implementation reduces operational costs for manual document review by 40–60%. For an average business, this saves tens of millions of rubles annually. P99 latency per document — under 200 ms, throughput — up to 50 documents per second. Our certified experience (10+ projects in fintech and government) guarantees results.

A CISO of one bank noted: "After implementing the system, we reduced the verification team by 30% and improved check quality — not a single fraudulent operation with forged documents in six months."

Implementation Results

In one project for a large bank, the system processes 50,000 documents per day, detecting 99% of forgeries before the manual verification stage. Document check time dropped from 2 minutes to 10 seconds. Operational cost savings amounted to tens of millions of rubles annually.

What the Work Includes

• Analysis of requirements and document types the business works with.
• Collection and labeling of a dataset: real documents, forged samples.
• Selection and training of models: from simple detectors to CNN ensembles.
• Integration via REST API with queue and caching support.
• Deployment in cloud (AWS, GCP) or on-premise.
• Documentation, employee training, granting model access.
• Post-release support: metric monitoring, retraining when new fraud types emerge.

Work Process

1. Analytics — we study your KYC process, document types, speed and accuracy requirements.
2. Design — we select architecture: stack, models, metrics.
3. Dataset — we label examples of forgeries and clean documents.
4. Training — we train and validate models, optimize latency.
5. Testing — we run A/B tests on your production traffic.
6. Deployment — we deploy, set up monitoring, train operators.

Additional information about methods can be found at Error level analysis and Convolutional neural network. Contact us — we will evaluate your project and offer an 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.