TEODY-Vision: A Deterministic, Non-Neural Framework for High-Speed UI Automation via Anisotropic Coherence and Multi-Scale Gradient Pyramids

Author: Teody A. Banagan
Affiliation: University of San Carlos & Concentrix Corporation

Abstract

Modern Robotic Process Automation (RPA) and graphical user interface (GUI) navigation heavily rely on resource-intensive neural networks or large Vision-Language Models (VLMs). While flexible, these architectures introduce computational latency, non-deterministic inference behaviors, and heavy hardware footprints. TEODY-Vision is a fully deterministic, non-neural framework for high-speed UI element localization and text tracking. By combining structure tensors, anisotropic coherence ($\Psi$), multi-scale gradient magnitude pyramids, and dynamic template synthesis, TEODY-Vision achieves sub-pixel localization accuracy with minimal CPU overhead.

Core Features

• Deterministic Execution: 100% execution repeatability across identical display buffers without probabilistic drift.
• Non-Neural & Lightweight: Zero dependency on heavy pre-trained weights or GPU accelerators; optimized for pure CPU multi-core execution.
• Dynamic Template Generation: Runtime synthetic template generation (generate_text_template), removing external static image dependencies.
• High-Speed Screen Capture: High-performance display buffer ingestion via mss.

Installation

pip install teody-vision


### Quick Start Code Block

Python import cv2 import numpy as np from mss import MSS from teody_vision import TEODYEngine

Capture monitor screen buffer

with MSS() as sct: monitor = sct.monitors[1] screenshot = sct.grab(monitor) frame = np.array(screenshot) frame = cv2.cvtColor(frame, cv2.COLOR_BGRA2BGR)

Initialize TEODY-Vision Engine

engine = TEODYEngine()

Run the tensor coherence and topological extraction pipeline

coherence, regions = engine.process_frame(frame)

print(f"[+] Isolated {len(regions)} topological UI regions successfully.")

## ⚖️ Legal Manifest & Copyright Notice

* **Official Registration:** Prepared for filing under the Intellectual Property Office of the Philippines (IPOPHL) within the Computer Programs / Software category.
* **Copyright Notice:** `Copyright (c) 2026 Teody Arboneda Banagan. All rights reserved.`
* **Open Source License:** Distributed globally under the highly permissive **MIT License**.

---

## 🔬 About the Author & Project Maintainer

**Teody Arboneda Banagan, PhD (Candidate)**
*Solution Architect, AI Engineer and Researcher*
* **Email:** teody882011@gmail.com

### Professional Stature & Domain Expertise
The structural maintenance, systems design, and algorithmic orchestration of this framework are driven by 20 years of progressive IT experience and top-tier credentials spanning international enterprise teams:
* **Doctor of Philosophy (PhD) in IT Candidate** — University of San Carlos, specializing in Advanced IT Frameworks, Robotics Process Automation, Computer Vision, and Enterprise AI Architectures.
* **Master in Information Technology (MIT)** — Cebu Institute of Technology - University.
* **Peer-Reviewed Reviewer & Author** — Invited Peer Reviewer for Scopus-indexed and Web of Science international journals.