Efficient Multi-Crop Saliency Partitioning for Automatic Image Cropping

2025-06-28

AI Summary

Traditional automatic image cropping methods generate only a single salient region, failing to support non-overlapping multi-region cropping. To address this limitation, this paper proposes the first efficient algorithm specifically designed for multi-cropping scenarios. Methodologically, it generalizes fixed-aspect-ratio single-bounding-box cropping to multi-region generation by dynamically adjusting attention thresholds and incrementally updating the saliency map—enabling non-overlapping region partitioning without redundant full-image saliency computation. The approach is theoretically guaranteed to achieve linear time complexity. Experiments demonstrate that the method effectively preserves key visual elements and compositional structure, significantly improving multi-cropping quality while maintaining high computational efficiency. This work establishes a novel paradigm for multi-region adaptive cropping and lays the foundation for future benchmark dataset construction and evaluation framework design in this domain.

Technology Category

• Computer Vision: Multi-modal Vision
• Machine Learning: Multi-instance/Multi-view Learning
• Intelligent Robots: Multimodal Perception & Sensor Fusion

Application Category

• Search and Retrieval-Augmented AI: Retrieval-Augmented Generation (RAG) and multi-modal RAG
• Graph Algorithms and Modeling for the Web: Efficient manipulation of static and dynamic Web-related graphs
• Web Mining and Content Analysis: Robustness and generalizability of Web mining methods

Abstract

Automatic image cropping aims to extract the most visually salient regions while preserving essential composition elements. Traditional saliency-aware cropping methods optimize a single bounding box, making them ineffective for applications requiring multiple disjoint crops. In this work, we extend the Fixed Aspect Ratio Cropping algorithm to efficiently extract multiple non-overlapping crops in linear time. Our approach dynamically adjusts attention thresholds and removes selected crops from consideration without recomputing the entire saliency map. We discuss qualitative results and introduce the potential for future datasets and benchmarks.

Problem

• Extracting multiple visually salient regions efficiently
• Avoiding recomputation of saliency map for disjoint crops
• Enhancing automatic image cropping with dynamic adjustments

Innovation

• Extends Fixed Aspect Ratio Cropping algorithm
• Dynamically adjusts attention thresholds
• Removes selected crops without recomputing