Enhancing the Interior Design Process through Data-Driven Aesthetic Assessment Systems

다국어 초록 (Multilingual Abstract)

Interior design is a multifaceted field that blends elements such as space, furnishings, and décor. To address the challenges of quantifying and enhancing this complex domain, our research leverages advanced artificial intelligence technologies. We introduce and evaluate three technology-driven methodologies aimed at refining and supporting the interior design process through quantifiable insights. The first study focuses on quantifying style elements and improving design communication between designers and non-expert users. By analyzing a large dataset of living room images, we generated data-driven furnishing pairing recommendations tailored to various interior styles. These pairing rules, based on crowd preferences, reflect the crowd's aesthetic assessment criteria and were validated through expert interviews, aiding communication and understanding of popular interior design styles. The second study develops the CMLsearch system for home décor product searches. This system quantifies interior design elements such as color, material, and lighting, allowing users to search for products that consider their existing environment. By supporting aesthetically consistent and harmonious choices, CMLsearch enhances usability and decision-making in product search and purchase processes, thereby supporting aesthetic assessment. The third study introduces the ICG system, which uses a Vision-Language Augmented Image Color Aesthetic Assessment model trained on a large dataset of crowd user preferences. This system assesses interior design images using the Mean Opinion Score and the Color Harmony Index, generating and evaluating aesthetically pleasing color-object combinations. The ICG system supports design creativity and evaluation by providing tools for aesthetic-aware design generation and assessment. Collectively, these methodologies enhance design communication, support aesthetically consistent decision-making, and foster creativity in the interior design process.

목차 (Table of Contents)

• CONTENTS i
• LIST OF FIGURES vi
• LIST OF TABLES ix
• ABSTRACT x
• CHAPTER 1. INTRODUCTION 1
• 1.1 Background 1
• 1.2 Objective 2
• 1.2.1 Enhancing Design Communication with Style-Specific Furnishing Pairing Rules Based on Crowd Preferences 3
• 1.2.2 Supporting Interior Products Exploration and Purchase with Semantic Visual Search System 4
• 1.2.3 Developing Crowd-Powered Aesthetic Assessment Model for Interior Design 8
• CHAPTER 2. RELATED WORK 9
• 2.1 Enhancing the Interior Design Process with Crowdsourced Furnishing Pairing Recommendations 9
• 2.1.1 Quantifying Design Style 9
• 2.1.2 Analyzing Furnishing and Pairing 11
• 2.1.3 Furnishing Pairing and Recommendation Methods 12
• 2.2 CMLsearch: Semantic Visual Search through Segmented Color, Material, and Lighting in Image 14
• 2.2.1 Search Intent and Interaction in Visual Search 14
• 2.2.2 Home Decor Product Searching and Pairing 15
• 2.2.3 Home Decor Product’s Attributes from Interior Image 17
• 2.3 Explanatory Feedback through Vision-Language Augmented Image Color Aesthetic Assessment Model for Interior Image Color Combination Generation 19
• 2.3.1 Color Design Support Tools 19
• 2.3.2 Computational Perceptual Model for Aesthetic Assessment 20
• 2.3.3 Exploratory feedback through Crowdsourced Ratings 20
• CHAPTER 3. ENHANCING THE INTERIOR DESIGN PROCESS WITH CROWDSOURCED FURNISHING PAIRING RECOMMENDATIONS 22
• 3.1 Methods 22
• 3.1.1 Data Collection and Processing 22
• 3.1.2 Adjusted Association Rule Mining 25
• 3.2 Adjusted Association Rule Results 28
• 3.2.1 Data Overview 28
• 3.2.2 Rule Generation with Adjusted Support 29
• 3.2.3 Rule Generation with Adjusted Confidence 31
• 3.2.4 Rule Generation with Adjusted Lift 33
• 3.3 Experts Interviews and Discussions 35
• 3.3.1 Interview Settings 35
• 3.3.2 Theoretical Implications 36
• 3.3.3 Managerial and Practical Implications 37
• 3.3.4 Limitations and Future Work 39
• 3.4 Chapter Summary 40
• 3.4.1Sub-Conclusion 40
• CHAPTER 4. CMLSEARCH: SEMANTIC VISUAL SEARCH THROUGH SEGMENTED COLOR, MATERIAL, AND LIGHTING IN IMAGE 41
• 4.1 CMLsearch 41
• 4.1.1 Segment-based CML Extraction Pipeline 41
• 4.1.2 Segment-based Object Similarity Calculation 46
• 4.1.3 Targeted CCT Adjustment for White Balance in Images 48
• 4.1.4 Analysis Metrics of User Behavior for Purchasing Home Decor Product 48
• 4.1.5 System Interface 52
• 4.2 User Study 56
• 4.2.1 Experimental Design 56
• 4.3 Results and Discussions 58
• 4.3.1 User Behavior in Target-Finding Scenario 59
• 4.3.2 User Behavior in Decision-Making Scenario 61
• 4.3.3 Key Features Compared to Conventional System 64
• 4.4 Chapter Summary 66
• 4.4.1 Sub-Conclusions 66
• CHAPTER 5. EXPLANATORY FEEDBACK THROUGH VISION-LANGUAGE AUGMENTED IMAGE COLOR AESTHETIC ASSESSMENT MODEL FOR INTERIOR IMAGE COLOR COMBINATION GENERATION 69
• 5.1 Vision-Language Augmented Image Color Aesthetic Assessment Model 69
• 5.1.1 Dataset 69
• 5.1.2 Model Architecture 70
• 5.1.3 Model Evaluation and Application 71
• 5.1.4 Visualization of Saliency Map 73
• 5.2 ICGsystem: Interior Image Color Combination Generation 74
• 5.2.1 System Overview and User Scenario 74
• 5.2.2 Image Color Parameters 75
• 5.2.3 Interaction Using Explanatory Feedback and Auto-Loop 76
• 5.2.4 Bayesian Optimization for Preferential Image Color Generation 77
• 5.3 User Study 78
• 5.3.1 Participants and Settings 78
• 5.3.2 Design Task and Procedure 79
• 5.4 User Study Results 80
• 5.4.1 Designers' Assessment of Color Harmony Index and MOS 80
• 5.4.2 Impact of Explanatory Feedback Through Saliency Maps on the Design Process 81
• 5.4.3 Impact of Auto-Loop Generated Color Combinations on the Design Process 83
• 5.4.4 Potential Application of the ICG System in Actual Design Processes for Interior Color Design 84
• 5.5 Discussions 85
• 5.5.1 Need for Multiple Assessment Tools 85
• 5.5.2 Impact of VL-ICAA Model Explainability on Designers in Design Contexts 85
• 5.6 Chapter Summary 86
• 5.6.1 Sub-Conclusions 86
• CHAPTER 6. CONCLUSION 87
• 6.1 Summary of Contributions 87
• 6.2 Future Work 88
• BIBLIOGRAPHY 90
• APPENDIX 106
• 국문요지 119
• ACKNOWLEDGMENTS 120
• CURRICULUM VITAE 121