File size: 3,372 Bytes
6731f94 061a1c3 616fad3 061a1c3 6731f94 061a1c3 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 |
# βοΈ Handwritten Signature Classifier
### AI-powered handwritten signature recognition for document processing
An image-classification model designed to distinguish handwritten signatures from non-signature content in scanned documents and document images.
Built on the **MobileNetV3-Small** architecture, this model is intended for efficient inference and integration into automated document-processing workflows.
---
## π Overview
The model classifies an input image into one of two categories:
| Class | Description |
|---|---|
| `SIGNATURE` | Handwritten signature or signature-like handwriting |
| `NOT_SIGNATURE` | Non-signature document content, such as printed text, dates, numbers, and other visual elements |
Its primary purpose is to help document-processing systems identify genuine signature regions and reduce false-positive detections.
## π§ Model Architecture
- **Architecture:** MobileNetV3-Small
- **Base Model:** `qualcomm/MobileNet-v3-Small`
- **Task:** Binary image classification
- **Framework:** Transformers
- **Input:** RGB image
- **Output:** Class probabilities and predicted label
- **Classes:** SIGNATURE / NOT_SIGNATURE
MobileNetV3-Small provides a lightweight architecture suitable for resource-constrained environments and desktop inference.
## π Training Dataset
**Dataset:** [`AmjadKha/HandWrittenSigntureDataset`](https://huggingface.co/datasets/AmjadKha/HandWrittenSigntureDataset)
The model uses a handwritten signature dataset for training and evaluation.
Supported document languages:
- English
- Arabic
## π Evaluation Metrics
The model is evaluated using the following metrics:
| Metric | Purpose |
|---|---|
| Accuracy | Measures overall classification correctness |
| Precision | Measures how many predicted signatures are actually signatures |
| Recall | Measures how many actual signatures are detected |
| F1-score | Balances precision and recall |
> Actual evaluation scores should be added after testing on a separate held-out test dataset.
## βοΈ Installation
```bash
pip install transformers torch pillow
```
## π» Usage
```python
from transformers import pipeline
classifier = pipeline(
"image-classification",
model="AmjadKha/YOUR-MODEL-REPOSITORY"
)
results = classifier("signature.png")
for result in results:
print(result["label"], result["score"])
```
Replace `YOUR-MODEL-REPOSITORY` with the actual Hugging Face model repository name.
## π¦ Intended Applications
- Banking document processing
- Signature verification workflows
- Account-opening documentation
- Customer onboarding documents
- Automated document classification
- PDF signature-region filtering
- Intelligent document processing (IDP)
## β οΈ Limitations
- Classification does not establish the legal authenticity of a signature.
- Handwritten names and initials may resemble signatures.
- Performance depends on image quality, resolution, and training-data diversity.
- The model should be evaluated on representative documents before production deployment.
- It should be combined with signature detection and verification mechanisms for high-assurance workflows.
## π License
Apache License 2.0.
## π¨βπ» Author
**Amjad Khaliliah**
Hugging Face: [@AmjadKha](https://huggingface.co/AmjadKha)
---
*Lightweight computer vision for intelligent document processing.*
|