MESA Lab researchers built and iterated a Scapy-based website-fingerprinting classifier that identifies specific YouTube videos from encrypted traffic by matching the packet-size sequence of thumbnail-image requests (triggered on SNI yt3.googleusercontent.com / yt3.ggpht.com) against a pre-built reference dataset (dataset_24_youtube_fingerprints.csv), scoring candidate matches on sequence overlap and total-payload-size ratio against fixed thresholds (0.65 / 0.85).
if 'yt3.googleusercontent.com' in CHO_payload or 'yt3.ggpht.com' in CHO_payload: test_s = generate_fingerprint(flows[flow_key]); handle_test_sample(test_s)
Defense implications
- Padding or randomizing the size/count of small auxiliary sub-resource fetches (thumbnails, favicons, etc.) that accompany a page load defeats this class of image-count/size-sequence fingerprinting.
- A transport that reorders or pads sub-resource fetch timing so it no longer mirrors the underlying page's fixed structure blunts classifiers built on stable per-page packet-size sequences like this one.
Related findings
A 2023 master's thesis from the Institute of Information Engineering, Chinese Academy of Sciences (MESA Lab's parent institute) develops an online encrypted-video-stream title identification method using response-time distribution and chunk-fingerprint matching that survives CDN node switching, reporting 96.19% title-identification accuracy using only 6 observed video chunks, without decrypting traffic.
A master's thesis-in-progress supervised within this ecosystem (advisor 杨嵘, senior engineer) is explicitly tied to the national key R&D program "海量公害 网页、图片、视频流量识别技术" (massive nuisance webpage/image/video traffic identification technology) and a VoIP engineering project; it builds cross- platform (YouTube/Facebook/Bilibili) encrypted-video identification that matches a single per-video traffic fingerprint across different CDNs/resolutions/codecs via trend-similarity, fuzzy-search, and meta-learning matching, explicitly framed around detecting "非法视频" (illegal/harmful video) spread across platforms.
An internal "网站/应用资源测绘" (website/app resource-mapping) presentation describes active EDNS-Client-Subnet DNS probing from many simulated geographic vantage points to map Google's and Facebook's global service-IP distribution by country/province, a mobile-app pipeline combining UI automation with MITM-proxy/ SSL-pinning-bypass to extract button-to-URL mappings from decrypted app traffic, and a "网站指纹审查" section that explicitly documents domain fronting as an effective evasion of its own DNS/SNI/Host-based website-fingerprint detection, alongside literature-based behavioral/ML website-fingerprinting intended to survive the QUIC/ECH/DoH transition.
Two IIE graduate-research repos (cuiyiming/gradproj, a 2019-2020 master's thesis project citing NDSS'17 TLS-interception-measurement and TLS-client-identification papers; daxiaoxu/xmr_bsexpr2, a 2022 project with GRU-based sequence classifiers over TCP/DNS flow JSON and deleted docs on TLS1.3 and Tencent's proprietary MMTLS protocol) document the ML feature-engineering methodology (TLS certificate length, handshake message sequences, JA3-style statistics, Markov-chain packet-size/timing models, GRU sequence models) that plausibly underlies production classifiers (e.g. stellar's later JA4/JA4S support, MESA_sts's randomness checks).
A MESA research-log entry details a NAT/shared-connection identification methodology combining TCP/IP fingerprinting (IP-ID, TTL, DF, window size, MSS, TCP-option ordering -- p0f-style), HTTP User-Agent diversity, TLS/SSL JA3 fingerprint diversity, and traffic statistical features (concurrent-TCP-connection count, idle-time jitter, upstream/downstream ratio stability, DNS query frequency) per endpoint over rolling time windows, with detection methods spanning direct UA inspection, threshold statistics, ML classifiers (random forest/SVM), and per-window entropy jumps across the fingerprint features.
A MESA Lab git repo (wangmeiqi/obfs4_meek_snowflake) trains closed-world Deep Fingerprinting (DF) website-fingerprinting classifiers (ClosedWorld_DF_NoDef.py) using pre-trained Keras models specifically for Tor's obfs4 pluggable transport, meek domain-fronting, and Snowflake -- the same transport families Lantern's own circumvention stack draws on.