Facial Recognition in Indian Criminal Investigations
How NCRB's CrPI and AI Are Transforming Modern Policing — 2026 Guide
Table of Contents
- Introduction: A Face in the Crowd, A Name in the Database
- How Facial Recognition Actually Works
- Delhi Protest Investigation (July 2026)
- NCRB's CrPI: India's New Era of Biometric ID
- Timeline: How India Got Here
- Comparison Tables
- Real-World Applications
- The Legal Framework
- Challenges and Limitations
- Case Studies
- Expert Insights
- Did You Know?
- Myth vs Reality
- The Future of Facial Recognition in India
- Key Points at a Glance
- Frequently Asked Questions
- Key Takeaways
- Conclusion
- References
Introduction: A Face in the Crowd, A Name in the Database
Somewhere in a Delhi police control room, a bank of monitors watches an intersection. A man in a hurry, face half-turned, passes under a streetlight. Within seconds, an algorithm has drawn a box around his face, compared it against a database of hundreds of thousands of faces, and returned a ranked list of possible matches. No human eye could do this at this speed, across this many cameras, for this many hours a day. That is the quiet, unglamorous reality of facial recognition in Indian policing today — not a Hollywood chase sequence, but rows of code comparing pixels to pixels, patiently, tirelessly, at scale.
2026 has been the year this technology moved from pilot projects to national infrastructure. In June, the Union Home Minister stood at the 26th All India Fingerprint Conference in New Delhi and launched four NCRB applications in one sweep — NCRB-Abhigyan, CrPI, e-Prosecution 2.0 and e-Forensics 2.0 — describing them as the next stage in India's push to connect an FIR to a conviction within three years (New Delhi Times, 2026). Weeks later, in the same city, facial recognition and AI-enabled CCTV analytics were at the centre of one of the most closely watched civil-liberties disputes of the year, as protesters at Jantar Mantar challenged the Delhi Police's surveillance practices before the Delhi High Court (Bar and Bench, 2026).
Both stories are really the same story. Facial recognition in India is simultaneously a genuine forensic breakthrough — capable of identifying missing children, tracing repeat offenders, and cutting the time it takes to solve certain categories of crime — and a technology whose legal guardrails are still being built in real time, argued out in courtrooms even as it is deployed on the street. This guide walks through both halves of that story: the science, the systems, the law, and the live controversy, without losing sight of the fact that a face is not a fingerprint, and a "match" is never, by itself, proof of guilt.
How Facial Recognition Actually Works — The Science, Explained Simply
Facial recognition sounds futuristic, but it is built from a sequence of fairly intuitive steps. Understanding each one matters, because every step is also a place where the system can go wrong.
Face Detection
Before a face can be recognised, it must first be found. Face detection algorithms scan an image or video frame and draw a bounding box around anything that looks like a human face — regardless of who it belongs to. This is the same basic technology that draws a little square around your face in a smartphone camera app.
Face Alignment
A detected face is rarely looking straight at the camera. Alignment software digitally rotates and warps the face so that the eyes, nose and mouth sit in a standard, predictable position — the way a passport photo studio asks you to look directly ahead and keep a neutral expression.
Landmark Detection
The system then plots dozens of specific points on the face — the corners of the eyes, the tip of the nose, the edges of the lips, the curve of the jaw. These landmarks act like a coordinate map of an individual face.
Feature Extraction and Face Embedding
This is where deep learning takes over. A neural network — usually a convolutional neural network trained on millions of face images — converts the landmark map and pixel patterns into a long string of numbers, typically a few hundred values, known as a "face embedding" or "faceprint." Two photos of the same person, even taken years apart, should produce embeddings that sit close together in this numerical space; two photos of different people should sit far apart.
Neural Networks and Deep Learning
The neural network that produces this embedding is trained by showing it enormous numbers of labelled face pairs — "these two photos are the same person," "these two are not" — until it learns which visual patterns reliably distinguish one face from another, independent of lighting, angle, or a change of hairstyle.
Similarity Score and Confidence Score
When an investigator submits a probe image — say, a CCTV still of a suspect — the system compares its embedding against every embedding in the reference database and returns a similarity score for each comparison, essentially a measure of numerical closeness. A confidence score expresses how sure the system is about the top matches, based on how much they stand out from the rest of the field.
False Positives and False Negatives
A false positive occurs when the system confidently matches two different people as the same person. A false negative occurs when it fails to match two photos of the same person. Both error types rise sharply with poor image quality, extreme angles, low light, ageing gaps between photos, and — as global research has repeatedly shown — vary depending on demographic factors such as skin tone and gender, a problem discussed further in the challenges section below.
Human Verification
Because of these error risks, no credible facial recognition deployment in Indian policing is supposed to end at the algorithm. A trained forensic examiner or officer is expected to visually review the top candidate matches, cross-check against other evidence such as fingerprints, call records, or eyewitness testimony, and only then treat identification as reliable enough to act on. This human-in-the-loop step is precisely where the July 2026 Delhi controversy discussed below becomes so important — because it is the step that critics say is at risk of being skipped or under-resourced.
Workflow: The Facial Recognition Process
- Image Capture — CCTV footage, mobile video, or a photograph is collected from a crime scene or public camera network.
- Face Detection — the software locates every face present in the frame.
- Preprocessing — poor-quality frames are enhanced (brightness correction, deblurring, cropping).
- Alignment & Landmarking — each face is standardised and mapped with key facial points.
- Embedding Generation — a neural network converts the face into a numerical faceprint.
- Database Comparison — the faceprint is compared against a reference database.
- Ranked Match List — the system returns candidate matches with similarity/confidence scores.
- Human Verification — a forensic expert or officer reviews the top matches manually.
- Corroboration — cross-checked against fingerprints, call records, eyewitnesses, or other evidence.
- Investigative Action — only after corroboration does the match feed into notices, summons, or arrest procedures, subject to legal safeguards.
Delhi Protest Investigation (July 2026): Facial Recognition in Real Criminal Investigation
Incident Overview
Since 20 June 2026, a sit-in protest and hunger strike organised under the banner of the satirically named "Cockroach Janta Party" (CJP), led by student activist Abhijeet Dipke, has been underway at Jantar Mantar in New Delhi. The protest was raised over repeated competitive examination paper leaks and demanded the resignation of the Union Education Minister (India Legal, 2026). The demonstration remained largely peaceful through its early weeks, but by 20–22 July 2026, clashes broke out between protesters, hardline agitators reported to have joined the gathering, and Delhi Police and Rapid Action Force personnel, resulting in injuries to security personnel and damage to public property (Organiser, 2026; Human Rights Watch, 2026).
Nature of the Protest and Reported Violence
Human Rights Watch reported that Indian security forces used tear gas and batons against what it described as largely peaceful student and youth protesters on 20 July 2026, and said it had verified video showing men in plain clothes joining uniformed personnel in beating demonstrators (Human Rights Watch, 2026). Separately, pro-government commentary and reports described the entry of non-student agitators into the protest and attacks on police personnel, framing the escalation as a law-and-order breakdown requiring firm investigation (Organiser, 2026). These are contested, partisan characterisations of the same events, and readers should treat both the "excessive force" framing and the "coordinated attack on police" framing as claims made by interested parties rather than settled fact pending independent inquiry.
Delhi Police's Investigation and the Use of AI-Assisted Facial Recognition
Following the clashes, Delhi Police stated that footage from its AI-enabled camera network was being used to identify people allegedly involved in the violence, with officials indicating that flagged individuals would be cross-checked against police databases before any arrests were made (Daily Pioneer, 2026). Ten FIRs were registered across multiple police stations in connection with the offences reported during the clashes (The News Minute, 2026). Delhi Police's wider surveillance network — expanded in phases to include roughly 10,000 additional cameras layered onto more than 15,000 previously integrated cameras — includes AI-powered video analytics capable of facial detection, automatic number-plate recognition, hand-gesture recognition and crowd-density monitoring (The News Minute, 2026).
AI-Enabled CCTV Analytics and Digital Evidence Collection
Investigative reporting has indicated that Delhi Police's facial recognition tools have been run against at least three distinct reference datasets: a database of more than 150,000 history-sheeters, a smaller database of around 2,000 individuals flagged as terror suspects, and a third dataset — described in reporting as containing "rabble-rousers and miscreants" — compiled from images gathered during past protests (The Indian Express, cited in The News Minute, 2026). A joint investigation by The Wire, conducted for the Pulitzer Center, separately reported that Delhi Police relied on facial recognition as the primary method for identifying individuals accused in earlier incidents of rioting (The News Minute, 2026).
Image Enhancement and Suspect Identification Workflow
As with earlier large-scale unrest in Delhi, the standard workflow for using facial recognition evidentially involves several careful steps: raw CCTV and mobile-phone footage submitted by the public and by police units is first catalogued; poor-quality frames are enhanced for brightness, focus and resolution; facial recognition software is then run to shortlist candidate matches from the reference databases; and, in past cases such as the 2022 Jahangirpuri violence investigation, arrested persons have even been re-photographed at a Forensic Science Laboratory in the same pose captured on CCTV, with court permission, to strengthen the evidentiary comparison (The Tribune, 2022). This illustrates that facial recognition matches are typically treated by Indian forensic practice as investigative leads to be strengthened with additional documentation, not as free-standing proof.
Human Verification Process and Legal Safeguards
Delhi Police, represented in court by the Solicitor General of India, has maintained before the Delhi High Court that its camera and recording operations at the protest site were conducted to maintain law and order, not to "snoop" on demonstrators, and denied that surveillance was disproportionate (Bar and Bench, 2026; India Legal, 2026). A Public Interest Litigation filed by former JNU Students' Union president Aishe Ghosh, however, alleges that Delhi Police has continuously photographed and videographed protesters — including during routine activities such as eating, resting, and seeking medical assistance — through a permanent surveillance tower installed at the protest site, and that no statutory basis, retention policy, or procedural safeguard for this monitoring has been disclosed (India Legal, 2026; NewsMobile, 2026). The Delhi High Court, hearing the matter, has asked whether any Standard Operating Procedure exists to regulate policing and surveillance at protest sites, and the matter remains listed for further hearing (The News Minute, 2026).
Why Facial Recognition Alone Cannot Establish Guilt
This case is a useful real-world illustration of a point forensic science teaches from its very first chapter: identification is not the same as proof. A facial recognition match tells an investigator "this person's face is numerically similar to a face in this photograph." It does not establish where that person was standing, what they did, what their intent was, or whether the underlying image itself was correctly time-stamped, unaltered, and lawfully obtained. Multiple technology-policy analysts have noted that combining biometric surveillance with an unclear legal basis, particularly at a site of political protest, raises the risk that identification technology could be used to trace not just alleged rioters, but also lawful, peaceful demonstrators (TechPolicy.Press, 2026; amLegals, 2026).
The Importance of Corroborative Forensic Evidence
For exactly this reason, sound forensic and prosecutorial practice treats a facial recognition output as one input among several. Call detail records placing a phone at the location, eyewitness statements taken under proper procedure, recovered physical evidence, fingerprint or DNA corroboration, and a documented, tamper-evident digital forensic chain of custody for the CCTV or mobile footage itself are all necessary before a facial recognition lead can responsibly support an arrest, let alone a chargesheet. This is not a limitation unique to India; it mirrors long-standing global forensic guidance on the use of probabilistic biometric evidence in criminal proceedings.
NCRB's Criminal Procedure Identification (CrPI): India's New Era of Biometric Criminal Identification
What Is CrPI?
CrPI — Crime and Criminal Profiling Identification — is a multi-modal biometric platform developed by the National Crime Records Bureau that allows investigators to simultaneously match fingerprints, facial features, iris patterns and DNA profiles through a single system, moving Indian policing away from a historical reliance on fingerprint identification alone (the420.in, 2026; Budding Forensic Expert, 2026). The Union Home Minister formally launched CrPI, alongside NCRB-Abhigyan, e-Prosecution 2.0 and e-Forensics 2.0, at the 26th All India Fingerprint Conference held in New Delhi on 19–20 June 2026 (New Delhi Times, 2026; GS Times, 2026).
Objectives and Why It Was Developed
The stated purpose of CrPI is to raise the accuracy and speed of criminal identification by combining biometric modalities that each have different strengths and weaknesses — facial recognition works from a distance and without physical contact; fingerprints offer very high discriminating power at close range; iris recognition is highly stable over a person's lifetime; and DNA offers the strongest evidentiary weight for biological trace evidence (the420.in, 2026). Home Minister Amit Shah, addressing the conference, linked the new applications to the government's wider criminal-law reform agenda since 2019, describing the goal as delivering justice — from FIR to conviction — within three years, and urged scientific data collection, evidence preservation, and database security (New Delhi Times, 2026).
Launch Details and Agencies Involved
The four applications were launched jointly, reflecting the intent to connect the entire investigative-to-prosecutorial chain: NCRB-Abhigyan (mobile, field-level fingerprint verification against NAFIS), CrPI (multi-modal biometric matching), e-Forensics 2.0 (digitally linking forensic science laboratories nationwide) and e-Prosecution 2.0 (linking case files to prosecution workflows) (New Delhi Times, 2026; the420.in, 2026). The event was attended by senior officials including the Director of the Intelligence Bureau, the Director of NCRB, and the Director of the Central Forensic Science Laboratory (New Delhi Times, 2026). These platforms sit within NCRB's broader Inter-Operable Criminal Justice System (ICJS), which links police (CCTNS), courts (e-Courts), prisons (e-Prisons), forensic laboratories (e-Forensics) and prosecution (e-Prosecution) on a shared data architecture approved by the e-Committee of the Supreme Court of India (Ministry of Home Affairs, 2026).
Legal Basis and Integration with the Criminal Procedure (Identification) Act, 2022
CrPI's legal foundation rests on the Criminal Procedure (Identification) Act, 2022, which repealed the colonial-era Identification of Prisoners Act, 1920, and substantially widened the category of "measurements" police and prison authorities may lawfully collect from arrested and convicted persons to include modern biometric identifiers such as iris and retina scans, alongside fingerprints, footprints, photographs and biological samples (Vajiram & Ravi, 2025; Budding Forensic Expert, 2026). Some states, including Delhi and Rajasthan, began implementing biometric collection under the Act as early as March 2025, drawing in part on Section 35(3) of the Bharatiya Nagarik Suraksha Sanhita, 2023, which permits biometric collection through investigative notices in certain circumstances (Budding Forensic Expert, 2026). The Act also contains a data-destruction safeguard: records of persons who have not previously been convicted and who are released without trial, discharged, or acquitted must generally be destroyed unless a Magistrate or Court directs otherwise for recorded reasons, with NCRB responsible for the governing Standard Operating Procedures (Budding Forensic Expert, 2026).
Future Vision
NCRB has positioned itself as evolving "from a record-keeping institution into an intelligence-based crime prevention organisation," with CrPI, NAFIS and the Abhigyan mobile application expected to expand in coverage, dataset size, and integration with state police forces over the coming years (Budding Forensic Expert, 2026).
CrPI's Core Features, Explained One by One
Facial Recognition
Matches a probe face — typically from CCTV or a photograph — against reference databases of known offenders, missing persons, or unidentified bodies, using the embedding-based process described earlier in this guide.
Fingerprint Identification
Draws on NAFIS, the National Automated Fingerprint Identification System launched in 2022, which now holds fingerprint records for more than 1.3 crore accused persons, convicts and prison inmates (the420.in, 2026).
Iris Recognition
Captures the unique, stable pattern of the iris — the coloured ring around the pupil — which changes far less over a person's lifetime than facial features do, making it valuable for long-term identification.
DNA Integration
Links biological sample analysis, conducted at forensic science laboratories, into the same identification pipeline, supporting cases where facial or fingerprint evidence alone is unavailable or degraded.
Multi-Modal Biometrics
Instead of relying on a single identifier, CrPI is designed to weigh evidence from several biometric modalities together, reducing the chance that a single technology's weakness — say, a blurry CCTV face — sinks an entire identification effort.
Repeat Offender Detection
By checking new arrests or crime-scene evidence against the accumulated national database, CrPI is intended to flag whether a suspect has a prior criminal history recorded anywhere in India, not just within a single state's records.
Missing Person Identification
Facial recognition and biometric matching support efforts to identify missing children and adults by comparing photographs against police and other government-held reference images.
Unknown Dead Body Identification
Where fingerprints or facial features can still be recovered, biometric matching can help identify unclaimed or unidentified bodies against missing-persons and criminal databases.
Criminal Database Integration
CrPI is built to work within NCRB's existing data architecture, including CCTNS (Crime and Criminal Tracking Network and Systems), so that identification results connect directly to case records already in the system.
AI-Assisted Identification
Machine learning models underlie the facial matching, and increasingly the ranking and prioritisation of candidate matches, reducing the manual workload on forensic examiners while still requiring their sign-off.
National Biometric Search
Because NAFIS and CrPI are national platforms rather than state-siloed systems, a fingerprint or face captured in one state can, in principle, be checked against records generated anywhere else in the country.
Future Scalability
NCRB officials have indicated that the architecture is intended to expand as more states digitise records and adopt the Abhigyan and CrPI applications at the field level, gradually building toward a genuinely national, real-time biometric identification capability (the420.in, 2026).
Workflow: How CrPI Processes an Identification Request
- Capture — Field officer or forensic unit captures biometric data using a certified device or the NCRB-Abhigyan mobile app.
- Transmission — Data is transmitted securely to the relevant NCRB database (NAFIS, the CrPI facial/iris module, or forensic laboratories for DNA).
- Query — The system runs the query against national records: criminal history, missing persons, and unidentified-body records.
- Ranked Results — Candidate matches are returned, typically within seconds for fingerprints and slightly longer for multi-modal queries.
- Review — A trained officer or forensic expert reviews the results and determines next investigative steps.
- Sync — Confirmed identifications and case updates are synced back into CCTNS and e-Prosecution 2.0.
Timeline: How India Got Here
| Year | Milestone |
|---|---|
| 1920 | Identification of Prisoners Act enacted under British colonial rule, limited to fingerprints, footprints and photographs of certain convicted persons (Vajiram & Ravi, 2025). |
| 1986 | National Crime Records Bureau established under the Ministry of Home Affairs (Budding Forensic Expert, 2026). |
| 2019 | Delhi Police's Automated Facial Recognition System (AFRS), originally built to trace missing children, is used at a public political rally for the first time, drawing early privacy criticism (Al Jazeera, 2019). |
| 2020–2021 | Facial recognition is used extensively to identify accused persons in the north-east Delhi riots and the Republic Day tractor-rally violence (Hindustan Times/PressReader, 2021). |
| 2022 | The Criminal Procedure (Identification) Act, 2022 is enacted, repealing the 1920 Act; NAFIS is launched the same year (Vajiram & Ravi, 2025). |
| 2022 | Facial recognition helps identify accused in the Jahangirpuri violence case; arrested persons re-photographed at an FSL for CCTV comparison (The Tribune, 2022). |
| Mar 2025 | Delhi and Rajasthan become the first states to implement biometric collection under the CrPI Act (Budding Forensic Expert, 2026). |
| Nov 2025 | The Digital Personal Data Protection Rules, 2025 are notified (hyperverge.co, 2026). |
| 19–20 Jun 2026 | NCRB-Abhigyan, CrPI, e-Forensics 2.0 and e-Prosecution 2.0 launched at the 26th All India Fingerprint Conference (New Delhi Times, 2026). |
| 20 Jun 2026 → | CJP-led protest begins at Jantar Mantar over examination paper leaks (India Legal, 2026). |
| Jul 2026 | PIL alleging intrusive surveillance filed; clashes follow; Delhi Police uses AI cameras to identify accused (India Legal, 2026; Daily Pioneer, 2026; HRW, 2026). |
Comparison Tables
Facial Recognition vs Fingerprint Identification
| Parameter | Facial Recognition | Fingerprint Identification |
|---|---|---|
| Contact required | No — works remotely, even covertly | Yes — typically needs physical contact or a latent print |
| Stability over time | Changes with age, weight, facial hair, surgery | Extremely stable across a lifetime |
| Effect of occlusion | Severely affected by masks, sunglasses, poor angle | Not affected by facial coverings |
| Speed at scale | Can scan crowds and video streams in real time | ~35 seconds per field check via NCRB-Abhigyan (Budding Forensic Expert, 2026) |
| Evidentiary weight in court | Generally treated as investigative lead requiring corroboration | Long-established, high evidentiary weight when properly matched |
| Bias/error concerns | Documented accuracy variation across demographics and image quality | Lower demographic variation; errors mainly from print quality |
AFRS vs CrPI
| Parameter | AFRS (Delhi Police) | CrPI (NCRB) |
|---|---|---|
| Scope | City-level; originally built for missing-children tracing | National platform, multi-modal biometric matching |
| Modalities | Facial recognition only | Facial recognition, fingerprint, iris and DNA integration |
| Operating agency | Delhi Police | National Crime Records Bureau, Ministry of Home Affairs |
| Legal grounding | No dedicated statute governing facial recognition specifically (The News Minute, 2026) | Anchored in the Criminal Procedure (Identification) Act, 2022 |
| Launch | First public deployment reported in 2019 | Launched June 2026 |
Traditional Identification vs AI Identification
| Parameter | Traditional Identification | AI-Assisted Identification |
|---|---|---|
| Method | Manual eyewitness identification, physical fingerprint comparison by an examiner | Automated embedding comparison across large databases |
| Scale | Limited to a handful of suspects at a time | Can screen thousands of faces or prints in minutes |
| Consistency | Subject to examiner fatigue and individual variation | Consistent processing, but consistent algorithmic bias if present |
| Transparency | Reasoning can usually be explained step by step | Neural network reasoning is harder to fully explain ("black box" concern) |
Face Recognition vs Iris Recognition
| Parameter | Face Recognition | Iris Recognition |
|---|---|---|
| Capture distance | Can work from a distance, even covertly via CCTV | Requires close-range, cooperative capture |
| Long-term stability | Moderate — changes with age and appearance | Very high — pattern is largely stable for life |
| Use case fit | Crowd monitoring, CCTV investigation, missing persons | Controlled checkpoints, prison intake, high-security verification |
Real-World Applications Beyond the Crime Scene
- Crime Investigation: Matching CCTV or mobile-phone imagery of suspects against criminal databases to generate investigative leads.
- Missing Persons: Comparing photographs of found children and adults against databases of reported missing persons.
- Human Trafficking: Assisting in identifying trafficking victims and repeat offenders across state lines, where a national database is especially valuable.
- Terrorism Investigation: Screening against watchlists of known or suspected individuals, subject to strict legal safeguards given the seriousness of consequences.
- Cybercrime: Linking digital evidence — profile photos, video calls, uploaded images — to real-world identities during investigation.
- Border Security: Verifying traveller identity against watchlists and travel documents at international checkpoints.
- Smart Policing: Supporting real-time situational awareness for police control rooms monitoring large public events.
- Prison Management: Verifying inmate identity during intake, transfer, and release to prevent impersonation.
- Crowd Monitoring: Assessing density and movement patterns at large gatherings for public-safety planning — a use distinct from, but sometimes blurred with, targeted identification.
- Disaster Victim Identification: Supporting identification of victims of mass-casualty events where facial features remain intact but other records are unavailable.
The Legal Framework: Rights, Rules and Grey Zones
Criminal Procedure (Identification) Act, 2022
This Act is the primary statute enabling police and prison authorities to collect an expanded range of "measurements," including biometric data, from arrested, convicted, and in some circumstances non-convicted persons under magisterial orders, with data-destruction safeguards attached (Vajiram & Ravi, 2025).
Digital Personal Data Protection Act, 2023
The DPDP Act, 2023 provides India's first comprehensive framework for processing digital personal data, and its Rules — notified in November 2025 — created operational categories such as Significant Data Fiduciaries and set out consent, notice, and breach-reporting obligations (hyperverge.co, 2026; CookieYes, 2026). However, legal commentators have pointed out that the DPDP Act was not specifically designed to regulate facial recognition technology or the particular constitutional questions raised by state use of biometric surveillance, leaving an acknowledged gap between data-protection law and facial-recognition-specific regulation (amLegals, 2026).
Right to Privacy and the Puttaswamy Judgment
In Justice K.S. Puttaswamy (Retd.) v. Union of India (2017), a nine-judge Bench of the Supreme Court unanimously held that the right to privacy is a fundamental right protected under Article 21 of the Constitution (Legal Service India, 2026). This judgment set out a now-standard test for any state action restricting privacy: it must have a legal basis, pursue a legitimate aim, be necessary, and be proportionate to that aim — precisely the test invoked in the Delhi High Court PIL over CJP protest surveillance (India Legal, 2026).
Ethical AI, Human Rights and Surveillance Concerns
Civil liberties groups and digital rights researchers have repeatedly flagged that India currently has no dedicated statute governing facial recognition technology itself, leaving its use to be justified under general policing powers rather than a purpose-built legal framework with defined limits on retention, audit, and permissible use cases (The News Minute, 2026; amLegals, 2026). Human Rights Watch's 2026 statement on the Delhi protests specifically called for prompt, impartial investigation into the use of force and for adherence to Supreme Court directions on transparency around internet-shutdown and surveillance orders (Human Rights Watch, 2026).
Court Admissibility
Indian courts have generally treated biometric identification, including facial recognition, as admissible investigative evidence when properly documented and corroborated — as illustrated by the FSL re-photography procedure used in the Jahangirpuri case, carried out only after securing court permission (The Tribune, 2022). Courts have not, to date, treated an uncorroborated facial recognition match alone as sufficient to establish guilt.
Challenges and Limitations Nobody Should Ignore
- Privacy: Continuous facial recognition of public spaces, including protest sites, raises the risk of chilling lawful assembly and speech.
- Surveillance creep: Databases built for one purpose can, over time, be repurposed for broader monitoring (The News Minute, 2026).
- False positives: Wrongly matching an innocent person to a database entry can trigger unwarranted police attention or reputational harm.
- Algorithm bias and dataset bias: Systems trained on unrepresentative image sets have, globally, shown uneven accuracy across different skin tones and genders.
- Low-resolution CCTV: Many Indian public cameras, especially older municipal installations, capture footage well below the resolution needed for reliable matching.
- Lighting: Poor or uneven lighting significantly degrades matching accuracy.
- Masks and face coverings: Partial occlusion reduces the landmark data available for matching.
- Ageing: A significant gap between a reference photo and a probe image increases the chance of a false negative.
- Twins and close relatives: Facial recognition can struggle to distinguish between identical twins or very similar-looking relatives.
- Human oversight: Rushed or understaffed verification raises the risk of an unchecked false match acting on real consequences.
- Cybersecurity: Centralised biometric databases are high-value targets; biometric identifiers cannot be "reset" after a breach.
Case Studies
Delhi Police reported using facial recognition matched against criminal records and driving-licence photographs to help identify accused persons following communal violence in February 2020, alongside video analytics, geo-location and DNA fingerprinting (Hindustan Times/PressReader, 2021).
Delhi Police analysed thousands of video clips and photographs submitted by the public and gathered from CCTV using video analytics and facial recognition to identify individuals involved in violence during the farmers' tractor rally (Hindustan Times/PressReader, 2021).
Facial recognition helped identify a number of accused persons; arrested individuals were subsequently re-photographed at a Forensic Science Laboratory, with court permission, in poses matching CCTV footage to strengthen the identification for evidentiary purposes (The Tribune, 2022).
This ongoing case illustrates both the operational use of AI-enabled camera networks to investigate reported violence and the parallel legal scrutiny of how such surveillance is authorised, documented and limited (multiple sources, 2026).
Expert Insights
Legal researchers examining Delhi Police's use of facial recognition have argued that the absence of transparent rules governing how reference databases are built and maintained creates the risk that identification technology could extend beyond legally defined categories of offenders to broader groups, including those merely present at a protest (The News Minute, 2026, paraphrased).
Analysts tracking the 2026 protests have observed that combining biometric surveillance with network shutdowns represents a mutually reinforcing pair of tools for managing public dissent, and have called for clearer, purpose-built legal limits distinct from general data-protection law (TechPolicy.Press, 2026, paraphrased).
Addressing the 26th All India Fingerprint Conference, the Home Minister urged investigators to prioritise scientific data collection, evidence preservation and database security as the foundation for the new NCRB platforms, situating CrPI within a broader push to deliver justice from FIR to conviction within three years (New Delhi Times, 2026, paraphrased).
Did You Know?
- NAFIS, launched in 2022, now holds fingerprint records for more than 1.3 crore accused persons, convicts and prison inmates across India (the420.in, 2026).
- NCRB-Abhigyan can return a criminal-history result to a field officer in approximately 35 seconds using a certified portable fingerprint scanner connected to a smartphone (Budding Forensic Expert, 2026).
- The Criminal Procedure (Identification) Act, 2022 repealed a law that had governed Indian criminal identification since 1920 — for over a century (Vajiram & Ravi, 2025).
- Delhi Police's camera-based surveillance network has been expanded in phases to add roughly 10,000 new cameras on top of more than 15,000 already integrated (The News Minute, 2026).
- India still has no single, dedicated law specifically regulating facial recognition technology, even as its use expands across policing, aviation, and other sectors (The News Minute, 2026; amLegals, 2026).
Myth vs Reality
| Myth | Reality |
|---|---|
| "A facial recognition match is enough to arrest and convict someone." | A match is an investigative lead. Indian forensic and prosecutorial practice requires corroboration through other evidence before it can support an arrest or chargesheet. |
| "Facial recognition is 100% accurate." | Accuracy varies with image quality, lighting, angle, ageing, and database size; false positives and false negatives are well-documented risks industry-wide. |
| "CrPI and AFRS are the same system." | AFRS is a Delhi Police facial-recognition-only system; CrPI is a national, NCRB-run, multi-modal biometric platform combining face, fingerprint, iris and DNA. |
| "There is a comprehensive Indian law specifically regulating facial recognition." | India currently regulates biometric data collection through the Criminal Procedure (Identification) Act, 2022 and general data-protection law under the DPDP Act, 2023, but has no facial-recognition-specific statute. |
| "Facial recognition at a protest automatically means everyone present is being profiled as a criminal." | Police have stated such deployments are for identifying those involved in reported violence, cross-checked against databases before action; petitioners and rights groups dispute whether this is adequately limited and transparent — a matter presently before the courts. |
The Future of Facial Recognition in India
Several trends are likely to shape the next phase of biometric policing in India. Wider rollout of CrPI, NAFIS and NCRB-Abhigyan across states is expected as digitisation of police records deepens and the applications integrate more fully with CCTNS and other ICJS components (Budding Forensic Expert, 2026). Smart-city camera networks, drone-based surveillance for crowd and traffic monitoring, and live CCTV analytics are expanding in parallel with facial recognition, raising the stakes of getting legal safeguards right. Edge AI — processing facial recognition directly on cameras rather than sending raw footage to central servers — is being explored globally as a way to reduce data-transmission risk, though its adoption in Indian policing remains at an early stage. Predictive policing tools, three-dimensional face recognition (which is more resistant to angle and lighting variation than 2D systems), and further multi-modal biometric fusion are all active areas of research and likely future NCRB initiatives, alongside continued legal and civil-society scrutiny of how these tools are governed.
Key Points at a Glance
- CrPI is a national, multi-modal biometric platform (face, fingerprint, iris, DNA) launched by NCRB in June 2026.
- It works alongside NAFIS (fingerprints), NCRB-Abhigyan (mobile field verification), e-Forensics 2.0 and e-Prosecution 2.0.
- Its legal foundation is the Criminal Procedure (Identification) Act, 2022, which replaced a 1920 colonial-era law.
- The July 2026 Delhi protest case shows both the investigative value and the contested legal boundaries of AI-enabled surveillance in real time.
- Facial recognition is an investigative lead, not standalone proof — corroboration remains essential.
- India lacks a dedicated facial-recognition-specific law; the DPDP Act, 2023 and CrPI Act, 2022 only partly fill this gap.
- Bias, false positives, poor image quality, and weak human oversight remain the technology's central risks.
Frequently Asked Questions
1. What is CrPI in simple terms?
CrPI is an NCRB software platform that lets police match a person's face, fingerprint, iris, or DNA profile against national criminal, missing-person, and other reference databases through one unified system.
2. When was CrPI launched?
CrPI was launched on 19–20 June 2026 at the 26th All India Fingerprint Conference in New Delhi, alongside three other NCRB applications (New Delhi Times, 2026).
3. Who developed CrPI?
CrPI was developed by the National Crime Records Bureau (NCRB), under the Ministry of Home Affairs, Government of India.
4. What is the difference between CrPI and NAFIS?
NAFIS is specifically a national fingerprint database and matching system, launched in 2022. CrPI is broader, combining facial recognition, fingerprint, iris and DNA matching in a single platform.
5. What is NCRB-Abhigyan?
NCRB-Abhigyan is a mobile application that lets field police officers verify fingerprints against NAFIS using a portable scanner connected to a smartphone, returning results in about 35 seconds (Budding Forensic Expert, 2026).
6. What is the Criminal Procedure (Identification) Act, 2022?
It is the law that repealed the 1920 Identification of Prisoners Act and expanded the biometric "measurements" — including iris and retina scans — that police and prison authorities may lawfully collect from arrested and convicted persons, with safeguards on data retention and destruction.
7. Is facial recognition legal in India?
There is no single, dedicated law specifically regulating facial recognition technology in India. Its use by police is generally justified under general policing powers and, where biometric data collection is involved, under the Criminal Procedure (Identification) Act, 2022, alongside the Digital Personal Data Protection Act, 2023 for data-processing obligations.
8. Can a facial recognition match alone lead to a conviction?
No. Indian forensic and prosecutorial practice treats a facial recognition match as an investigative lead requiring corroboration from other evidence, such as fingerprints, call records, or eyewitness testimony, before it can support a chargesheet or conviction.
9. What happened at the Delhi protest in July 2026?
A protest led by the Cockroach Janta Party at Jantar Mantar, ongoing since 20 June 2026 over examination paper leaks, saw clashes with police in July 2026. Delhi Police used AI-enabled camera analytics to identify individuals allegedly involved in the violence, while a Public Interest Litigation before the Delhi High Court separately challenged the broader surveillance of the protest itself.
10. What is the AFRS used by Delhi Police?
The Automated Facial Recognition System (AFRS) is Delhi Police's facial-recognition-only system, originally deployed to trace missing children before being used more broadly, including at public events and protests.
11. What is a false positive in facial recognition?
A false positive occurs when the system incorrectly matches two different people as being the same individual, which can lead to a wrongly flagged suspect if not caught through human verification.
12. What is a false negative in facial recognition?
A false negative occurs when the system fails to match two photographs of the same person, potentially causing an investigator to miss a legitimate identification.
13. Why can't facial recognition tell twins apart reliably?
Facial recognition relies on measurable visual features and proportions, which can be nearly identical between identical twins, unlike fingerprints or DNA, which differ even between twins.
14. Does facial recognition work with masks or face coverings?
Accuracy drops significantly with masks or heavy face coverings because key landmarks around the nose and mouth are hidden, reducing the data available for matching.
15. What is the Puttaswamy judgment and why does it matter here?
It is the 2017 Supreme Court ruling that recognised privacy as a fundamental right under Article 21, establishing a legality-necessity-proportionality test that is now central to legal challenges against surveillance practices, including facial recognition.
16. What does the Digital Personal Data Protection Act, 2023 say about facial data?
It sets out general obligations for processing personal data, including sensitive biometric information, but was not designed specifically to regulate facial recognition technology or its unique constitutional concerns.
17. How is DNA integrated into CrPI?
CrPI is designed to connect forensic laboratory DNA analysis into the same identification workflow as facial, fingerprint and iris data, supporting cases where other biometric evidence is degraded or unavailable.
18. What are the main criticisms of facial recognition in Indian policing?
Key criticisms include the absence of a dedicated legal framework, lack of transparency about how reference databases are built and audited, risk of surveillance extending to peaceful protesters, and documented accuracy limitations across image quality and demographic factors.
19. How does facial recognition help find missing persons?
Photographs of found individuals are compared against databases of persons reported missing, allowing investigators to flag potential matches for further verification, including in cases involving children.
20. What is e-Forensics 2.0?
e-Forensics 2.0 is one of the four NCRB applications launched in June 2026, designed to digitally connect forensic science laboratories across India to improve evidence tracking and information sharing.
21. What is e-Prosecution 2.0?
e-Prosecution 2.0 is an NCRB platform launched alongside CrPI that links investigation and forensic case data to prosecution workflows, aiming to reduce delays between investigation and trial.
22. Can facial recognition evidence be challenged in court?
Yes. Like other forensic evidence, the reliability of the underlying image, the chain of custody, the matching methodology, and the presence of corroborating evidence can all be challenged during trial.
Key Takeaways
- India's biometric policing infrastructure took a major leap forward in June 2026 with the launch of CrPI, NCRB-Abhigyan, e-Forensics 2.0 and e-Prosecution 2.0.
- CrPI's strength lies in combining facial, fingerprint, iris and DNA identification into a single national platform grounded in the Criminal Procedure (Identification) Act, 2022.
- The July 2026 Delhi protest controversy is a live, real-world test of how facial recognition and AI-enabled surveillance interact with constitutional rights to privacy, speech and assembly.
- Facial recognition is a powerful investigative tool but remains probabilistic — it identifies candidates, not certainties, and always requires human verification and independent corroboration.
- India's legal framework — the CrPI Act, 2022 and the DPDP Act, 2023 — regulates biometric data collection and processing generally, but a dedicated, facial-recognition-specific law remains absent.
Conclusion
Facial recognition has moved, in the space of a few years, from an experimental tool used to trace missing children to a central pillar of India's national criminal-identification architecture. CrPI represents genuine forensic progress — a single platform capable of weighing face, fingerprint, iris and DNA evidence together, built on top of NAFIS's already substantial fingerprint database and backed by a dedicated statute in the Criminal Procedure (Identification) Act, 2022. At the same time, the events at Jantar Mantar in July 2026 are a reminder that the same cameras and algorithms that help identify a missing child or a repeat offender can, without careful legal boundaries, be turned on peaceful citizens exercising their constitutional right to protest. The forensic science community's responsibility in this moment is not to choose a side in that debate, but to keep insisting on the discipline's oldest lesson: a match is a lead, not a verdict, and every biometric identification deserves the same rigorous, corroborated, transparent process that any other piece of forensic evidence would demand.
References
- amLegals. "Facial Recognition in India: Privacy, Surveillance and the Need for Regulation." amLegals, 2026. amlegals.com
- Bar and Bench. "Ensuring law and order, not snooping: Delhi Police on PIL in Delhi HC alleging surveillance of CJP protests." Bar and Bench, July 2026. barandbench.com
- Budding Forensic Expert. "CrPI — Crime & Criminal Profiling Identification." Budding Forensic Expert, June 2026. buddingforensicexpert.in
- Budding Forensic Expert. "NCRB-Abhigyan: India's Mobile Fingerprint Revolution — A Complete Guide." Budding Forensic Expert, June 2026. buddingforensicexpert.in
- CookieYes. "India Digital Personal Data Protection Act (DPDPA 2025): Updated Guide." CookieYes, January 2026. cookieyes.com
- Daily Pioneer. "Delhi Police Uses AI Surveillance to Identify Accused in CJP Protest Violence." Daily Pioneer, July 2026. dailypioneer.com
- Human Rights Watch. "India: Excessive Force Used Against Student Protesters." Human Rights Watch, July 23, 2026. hrw.org
- hyperverge.co. "The 2026 Guide to Facial Recognition and Privacy in India." hyperverge.co, June 2026. hyperverge.co
- India Legal. "Petition filed in Delhi High Court challenges surveillance of CJP protesters by Delhi Police." India Legal, July 2026. indialegallive.com
- India Legal. "Delhi Police to Delhi High Court: Surveillance of CJP protesters was for security, not snooping." India Legal, July 2026. indialegallive.com
- Legal Service India. "Digital Personal Data Protection Act, 2023." Legal Service India, April 2026. legalserviceindia.com
- New Delhi Times. "Amit Shah Inaugurates 26th All India Fingerprint Conference 2026, Launches Four NCRB Applications." New Delhi Times, June 20, 2026. newdelhitimes.com
- NewsMobile. "Delhi High Court Questions Police Over Surveillance At CJP Protest." NewsMobile, July 2026. newsmobile.in
- Ministry of Home Affairs, Government of India. "ICJS/NCRB Administration." mha.gov.in, March 2026. mha.gov.in
- Organiser. "Mob rule on Delhi's streets: Police face coordinated attack, vandalism and tear gas mark fresh clashes." Organiser, July 23, 2026. organiser.org
- PressReader/Hindustan Times. "Amid protests and riots, police put faith in tech." Hindustan Times via PressReader, February 2021. pressreader.com
- the420.in. "Government Launches Four Digital Platforms to Link Crime Investigation and Prosecution." the420.in, June 2026. the420.in
- The News Minute. "Explained: Delhi Police's AI surveillance network and why it's under scrutiny." The News Minute, July 2026. thenewsminute.com
- The Tribune. "Jahangirpuri violence: Delhi Police gets pics of accused clicked to match them with CCTV footage." The Tribune, August 11, 2022. tribuneindia.com
- TechPolicy.Press. "India's CJP Protests Meet Internet Shutdowns and Pervasive Surveillance." TechPolicy.Press, July 2026. techpolicy.press
- Vajiram & Ravi. "The Criminal Procedure (Identification) Act (CrPI): Challenges in Implementation and Way Ahead." Vajiram & Ravi, April 22, 2025. vajiramandravi.com
