Parameter-free structure-texture image decomposition by unrolling

2025-03-17

AI Summary

This work addresses the structure–texture image decomposition problem. We propose LPR-NET, the first fully parameter-free and computationally efficient learnable method for this task. Grounded in the low-patch-rank (LPR) prior, LPR-NET unrolls an optimization process into a sequence of trainable neural network layers, enabling end-to-end supervised learning. Unlike conventional iterative algorithms, it eliminates reliance on handcrafted priors and manually tuned hyperparameters. Trained solely on synthetic data, LPR-NET generalizes effectively to natural images without adaptation. Experiments demonstrate that LPR-NET achieves decomposition quality competitive with classical model-based methods, while offering significantly faster inference—orders of magnitude quicker than iterative solvers. Crucially, it requires zero human intervention for parameter tuning. To our knowledge, LPR-NET is the first approach to simultaneously achieve full parameter freedom and computational efficiency in structure–texture decomposition.

Technology Category

• Computer Vision: Learning & Optimization for CV
• Machine Learning: Structured Learning
• Natural Language Processing: Learning & Optimization for NLP

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Graph Algorithms and Modeling for the Web: Algorithms and analysis for incomplete, noisy, or partially observed Web-related graphs
• Economics, Online Markets and Human Computation: Cost models of using LLMs in production systems

Abstract

In this work, we propose a parameter-free and efficient method to tackle the structure-texture image decomposition problem. In particular, we present a neural network LPR-NET based on the unrolling of the Low Patch Rank model. On the one hand, this allows us to automatically learn parameters from data, and on the other hand to be computationally faster while obtaining qualitatively similar results compared to traditional iterative model-based methods. Moreover, despite being trained on synthetic images, numerical experiments show the ability of our network to generalize well when applied to natural images.

Problem

• Parameter-free method for structure-texture image decomposition
• Neural network LPR-NET based on Low Patch Rank model
• Generalizes well from synthetic to natural images

Innovation

• Parameter-free image decomposition method
• LPR-NET based on Low Patch Rank unrolling
• Generalizes well from synthetic to natural images