About CipherLens

A machine learning system for automated classical cipher identification.

Project Overview

Classical cipher identification is the critical first step in any cryptanalysis workflow. Traditional manual methods are time-consuming and expertise-dependent.

CipherLens automates this process using machine learning. Given only ciphertext (no plaintext or keys), our models extract 15 statistical features and classify the text into one of 22 cipher types across 6 cryptographic families.

The system offers three model engines: a Hybrid CNN (79.24% acc) combining character-level patterns with statistical features, a CNN Deep Learning model (68.47% acc) reading raw character sequences, and an XGBoost hierarchical classifier using a two-stage family → cipher pipeline with soft-routing.

How It Works
Ciphertext Input
User pastes encrypted text
Feature Extraction
15 statistical features computed
Classification
Hybrid CNN, DL, or XGBoost
Prediction
Top 3 ciphers with confidence
Supported Ciphers (22)
FamilyCiphersCount
Monoalphabetic SubstitutionCaesar, Affine, Atbash
3
Polyalphabetic SubstitutionVigenere, Autokey, Beaufort, Porta
4
TranspositionColumnar Transposition
1
Polygraphic SubstitutionPlayfair, Hill, Four-Square
3
FractionatingBifid, Trifid, ADFGX, ADFGVX, Nihilist, Polybius
6
Modern BlockLucifer, MISTY1, LOKI, TEA, XTEA
5
Total
22
Team
Dhruv Verma
2022172
Maulik Mahey
2022282
Md Kaif
2022289
Sweta Snigdha
2022527
Supervised by Dr. Ravi Anand — IIIT Delhi
BTP 2025-2026