ClassScan is a web-based attendance management system that uses face recognition to automatically identify students from classroom photos and generate attendance reports as downloadable CSV files.

Faculty takes a few photos of the class from different angles, uploads them to the system, and within minutes a complete attendance sheet — with every student marked present or absent — is ready to download.

• Multi-photo Registration — Students register using 5–10 photos for maximum recognition accuracy across different angles and lighting conditions
• Live Camera Capture — Students can take selfies directly in the browser without uploading files
• Batch Attendance Processing — Faculty uploads multiple class photos; results are merged automatically, eliminating duplicates
• Instant CSV Export — Attendance report downloads automatically with roll number, name, status, time, and date
• Student Dashboard — View all registered students with registration details
• Database Management — Clear student database or attendance records independently with confirmation prompts
• Fully Local — No cloud, no external APIs, no data leaves your machine

http://localhost:5000/           →  Home Dashboard
http://localhost:5000/register   →  Student Registration
http://localhost:5000/attendance →  Take Attendance
http://localhost:5000/students   →  View All Students

REGISTRATION
────────────
Student opens /register
       ↓
Takes 5–10 selfies via browser camera
       ↓
DeepFace generates face embeddings (128 numbers per photo)
       ↓
Embeddings are averaged → one unified face signature
       ↓
Saved to local JSON database


ATTENDANCE
──────────
Faculty uploads 3–5 class photos from different angles
       ↓
System detects all faces in each photo
       ↓
Each face is encoded → compared against database
       ↓
Closest match below threshold → marked PRESENT
No match found → marked ABSENT
       ↓
Results merged across all photos (no duplicates)
       ↓
CSV downloaded automatically

• Python 3.11
• pip
• Git
• A webcam or phone camera

1. Clone the repository

git clone https://github.com/yourusername/ClassScan.git
cd ClassScan

2. Create virtual environment

py -3.11 -m venv attendance_env
attendance_env\Scripts\activate        # Windows
source attendance_env/bin/activate     # Mac/Linux

3. Install dependencies

pip install -r requirements.txt

⚠️ First install may take 5–10 minutes. DeepFace downloads the Facenet model (~90MB) on first run.

4. Create required folders

mkdir dataset encodings attendance static\uploads templates

5. Run the app

python app.py

6. Open in browser

http://localhost:5000

1. Go to /register
2. Enter student name and roll number
3. Click Open Camera and take 5–10 selfies
4. Click Register Student
5. Student is now in the system

💡 More photos = better accuracy. Vary angles and expressions.

1. Faculty takes 3–5 photos of the classroom from different angles
2. Go to /attendance
3. Upload all class photos
4. Click Process Attendance
5. CSV downloads automatically

• View all registered students at /students
• Clear Database — removes all student registrations
• Clear Attendance — removes all generated CSV files

ClassScan/
│
├── app.py                  ← Flask web application
├── register.py             ← Student registration logic
├── recognize.py            ← Face recognition + attendance logic
├── utils.py                ← Shared database utilities
├── requirements.txt        ← Python dependencies
│
├── templates/
│   ├── index.html          ← Home page
│   ├── register.html       ← Registration form with camera
│   ├── attendance.html     ← Attendance upload form
│   └── students.html       ← Student management dashboard
│
├── static/
│   └── uploads/            ← Temporary uploaded files
│
├── dataset/                ← Student registration photos
│   └── {roll_no}/          ← One folder per student
│
├── encodings/
│   └── database.json       ← Face embeddings database
│
└── attendance/             ← Generated attendance CSV files

You can adjust recognition sensitivity in recognize.py:

# Lower = stricter matching (fewer false positives)
# Higher = more lenient matching (fewer missed detections)
THRESHOLD = 10

Recommended values:

• 8–10 — High accuracy environment (controlled lighting)
• 10–12 — Normal classroom environment
• 12–15 — Low light or varied angles

• All data is stored locally on your machine. Nothing is uploaded to any server.
• Student photos in dataset/ and encodings/database.json contain biometric data. Do not share or commit these files.
• The .gitignore excludes all sensitive data by default.

flask
deepface
opencv-python
numpy
tf-keras
werkzeug

MIT License — feel free to use, modify and distribute.