PRESS RELEASE:

JUNE 2026 

INVENTOR LEO PORTAL RELEASES TECHNICAL WHITE PAPER ON IDENTITY-PRESERVING AI IMAGE RESTORATION

Two issued U.S. patents underpin a new framework asking whether identity fidelity should become a verifiable per-output property of AI restoration systems.

Leo Portal, sole inventor of two issued U.S. patents relating to image identification technologies, has released a technical white paper examining how concepts from those patents may be interpreted within the context of modern artificial intelligence and generative image restoration.

White Paper Title:

Identity-Preserving AI Image Restoration Using Adaptive Feature Weighting and Multi-Image Template Generation

UNDERLYING INTELLECTUAL PROPERTY

• U.S. Patent No. 12,412,422 B2 — Method and System for Verifying Image Identification (Issued)

• U.S. Patent No. 099642-005827USC1 — Method and System for Verifying Image Identification - Infrared - (Issued) 

• U.S. Patent Publication No. 2018/0068173 A1 — Method and System for Verifying Image Identification (Published Application)

                                                                                      THE PROBLEM

Current AI restoration systems have largely solved the realism problem — a restored image can look beautiful and photorealistic while no longer accurately representing the original subject. Yet restoration models frequently alter facial proportions, introduce hallucinated features, or subtly change the identity of the subject being restored, a phenomenon known as identity drift. This gap between visual plausibility and identity fidelity remains an important and only partially solved problem. To address this, this paper extends methods introduced in previously issued image identification patents (U.S. Patents 12,412,422 B2 and 12,418,720 B2), proposing an Identity-Constrained Supervisory Framework (ICSF) that applies adaptive feature weighting and inference-time identity verification to existing AI restoration architectures — augmenting rather than replacing them.

THE FRAMEWORK:

The white paper proposes an identity-constrained architecture in which identity fidelity becomes a measurable, verifiable property of each restored image rather than merely an average tendency shaped during model training.

Instead of replacing diffusion and transformer architectures, the framework acts as a supervisory layer incorporating:

• Multi-image identity template generation

• Adaptive per-feature weighting

• Embedding-based similarity scoring

• Iterative inference-time feedback loops

• Identity drift detection and correction prior to image delivery

CENTRAL QUESTION:

Should identity fidelity be treated as an average characteristic learned during training, or should it become a measurable and verifiable property of every output generated during inference?

BRIDGING TWO RESEARCH COMMUNITIES:

The paper surveys both face restoration systems and identity-aware generation systems and proposes that combining the restoration capabilities of one with the identity verification rigor of the other represents a meaningful open research direction.

FACE RESTORATION SYSTEM:

• GFPGAN

• CodeFormer

• RestoreFormer

Identity-Aware Generation Systems

• DreamBooth

• InstantID

• PhotoMaker

• ConsisID

PAPER CONTENTS:

The white paper includes:

• Formal mathematical notation

• Literature review with citations

• Proposed benchmark methodology

• Human identity recognition metrics

• Failure-case analysis

• Acknowledged limitations

• Future research directions

POTENTIAL APPLICATIONS:

• Historical photo restoration

• Genealogy and archival preservation

• Forensic imaging

• Authentication systems

• Video restoration

• Personalized avatars

• Medical image reconstruction

• Multimodal identity embeddings

AVAILABILITY:

The complete white paper is available for download through this website.

MEDIA AND RESEARCH INQUIRIES:

Leo Portal

(Independent Inventor and Research Author)