Driven by the huge profit potential in the fashion industry, intelligent fashion analysis may become an important subject in the multimedia and computer vision research. Traditional vision-based clothing research methods focused on analyzing fashion items based on either keywords given by users or low-level features specified by preferred samples. Instead of using less-discriminative low-level features or ambiguous keywords to analyze fashion items, this study proposes novel approaches that focus on clothing genre recognition and fashion trends analysis based on the visually-differentiable fashion style elements. A set of style elements that are crucial for recognizing clothing genres and analyzing fashion trends are identified based on the fashion design theory. In addition, the corresponding salient visual features of each style element are identified and formulated with variables that can be computationally derived with various computer vision algorithms.

In terms of clothing genre recognition, we propose a novel classification technique to identify the genres of upperwear and lowerwear from full-body pictures through recognizing fundamental style elements of clothing design, such as collars, front buttons, and sleeves. We extract the representative features for describing style elements based on the spatial layout of body-parts. In addition, we make one step ahead to automatically classifying clothing genres by introducing the advantage of integrating local features of multimodality as the instances of prize-collecting Steiner tree (PCST) problem to discover clothing regions, and exploiting visual style elements to discover the clothing genre. Recognition results show that our clothing genre recognition frameworks have significant performance and superiority in comparison with the state-of-the-art recognition methods. Moreover, the effectiveness of each style element and its visual features on recognizing clothing genres are demonstrated through a set of experiments involving different sets of style elements or features.

On the topic of fashion trend spotting, we aim to present a novel algorithm that automatically discovers visual style elements representing fashion trends for a certain season of fashion week events. The five major elements of fashion style (i.e. head decoration, color, silhouette, pattern, and footwear) are investigated in this framework. The trending styles are discovered based on the stylistic coherent and unique characteristics of fashion style elements. The experimental evaluations and analysis on a large number of catwalk show videos well demonstrate the effectiveness of our proposed method.

Driven by the huge profit potential in the fashion industry, intelligent fashion analysis may become an important subject in the multimedia and computer vision research. Traditional vision-based clothing research methods focused on analyzing fashion items based on either keywords given by users or low-level features specified by preferred samples. Instead of using less-discriminative low-level features or ambiguous keywords to analyze fashion items, this study proposes novel approaches that focus on clothing genre recognition and fashion trends analysis based on the visually-differentiable fashion style elements. A set of style elements that are crucial for recognizing clothing genres and analyzing fashion trends are identified based on the fashion design theory. In addition, the corresponding salient visual features of each style element are identified and formulated with variables that can be computationally derived with various computer vision algorithms.

In terms of clothing genre recognition, we propose a novel classification technique to identify the genres of upperwear and lowerwear from full-body pictures through recognizing fundamental style elements of clothing design, such as collars, front buttons, and sleeves. We extract the representative features for describing style elements based on the spatial layout of body-parts. In addition, we make one step ahead to automatically classifying clothing genres by introducing the advantage of integrating local features of multimodality as the instances of prize-collecting Steiner tree (PCST) problem to discover clothing regions, and exploiting visual style elements to discover the clothing genre. Recognition results show that our clothing genre recognition frameworks have significant performance and superiority in comparison with the state-of-the-art recognition methods. Moreover, the effectiveness of each style element and its visual features on recognizing clothing genres are demonstrated through a set of experiments involving different sets of style elements or features.

On the topic of fashion trend spotting, we aim to present a novel algorithm that automatically discovers visual style elements representing fashion trends for a certain season of fashion week events. The five major elements of fashion style (i.e. head decoration, color, silhouette, pattern, and footwear) are investigated in this framework. The trending styles are discovered based on the stylistic coherent and unique characteristics of fashion style elements. The experimental evaluations and analysis on a large number of catwalk show videos well demonstrate the effectiveness of our proposed method.

Abstract . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . i
Acknowledgment . . . . . . . . . . . . . . . . . . . . . . . . . . .iii
Table of contents . . . . . . . . . . . . . . . . . . . . . . . . . . v
List of Tables . . . . . . . . . . . . . . . . . . . . . . . . . . viii
List of Figures . . . . . . . . . . . . . . . . . . . . . . . . . . .ix
1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . .1
1.1 Background on Fashion Analysis . . . . . . . . . . . . . . . . . .1
1.2 Contributions of the Dissertation . . . . . . . . . . . . . . . . 2
1.3 Organization of the Dissertation . . . . . . . . . . . . . . . . .3
2 Learning and Recognition of Clothing Genres from Full-body Images . 5
2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . .5
2.2 Related Work . . . . . . . . . . . . . . . . . . . . . . . . . . .6
2.3 System Overview . . . . . . . . . . . . . . . . . . . . . . . . .11
2.4 Style Elements of Clothes . . . . . . . . . . . . . . . . . . . .12
2.4.1 Style Elements of Upperwear . . . . . . . . . . . . . . . . . .12
2.4.2 Style Elements of Lowerwear . . . . . . . . . . . . . . . . . .20
2.5 Clothing Genre Learning and Classification . . . . . . . . . . . 28
2.6 Experiments . . . . . . . . . . . . . . . . . . . . . . . . . . .29
2.6.1 Dataset . . . . . . . . . . . . . . . . . . . . . . . . . . . .30
2.6.2 Experimental Setup . . . . . . . . . . . . . . . . . . . . . . 32
2.6.3 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . .33
2.7 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
3 Clothing Discovery in Consumer Photos . . . . . . . . . . . . . . .45
3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . 45
3.2 Related Work . . . . . . . . . . . . . . . . . . . . . . . . . . 47
3.2.1 Clothing Region Detection . . . . . . . . . . . . . . . . . . .48
3.2.2 Clothing Genre Recognition . . . . . . . . . . . . . . . . . . 49
3.3 Clothing Region Detection . . . . . . . . . . . . . . . . . . . .50
3.3.1 Offline-Trained Local Classifier . . . . . . . . . . . . . . . 51
3.3.2 Clothes Pixel Labelling . . . . . . . . . . . . . . . . . . . .52
3.4 Clothing Genre Recognition . . . . . . . . . . . . . . . . . . . 54
3.4.1 Clothing Style Elements . . . . . . . . . . . . . . . . . . . .56
3.4.2 Classification and Recognition . . . . . . . . . . . . . . . . 60
3.5 Experiments . . . . . . . . . . . . . . . . . . . . . . . . . . .61
3.5.1 Dataset . . . . . . . . . . . . . . . . . . . . . . . . . . . .61
3.5.2 Performance on Clothing Region Detection . . . . . . . . . . . 62
3.5.3 Performance on Clothing Genre Recognition . . . . . . . . . . .63
3.6 Conclusion and Future Work . . . . . . . . . . . . . . . . . . . 68
4 Fashion Trends Spotting by Exploiting the Style Elements . . . . . 70
4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . 70
4.2 Related Work . . . . . . . . . . . . . . . . . . . . . . . . . . 72
4.3 System Framework . . . . . . . . . . . . . . . . . . . . . . . . 73
4.3.1 Showcased Garment Identification . . . . . . . . . . . . . . . 74
4.3.2 Feature Extraction for Each Fashion Style Element . . . . . . .75
4.3.3 Trending Style Initialization . . . . . . . . . . . . . . . . .81
4.3.4 Trending Style Selection . . . . . . . . . . . . . . . . . . . 82
4.4 Experimental Results . . . . . . . . . . . . . . . . . . . . . . 83
4.4.1 Dataset . . . . . . . . . . . . . . . . . . . . . . . . . . . .83
4.4.2 Results and Discussion . . . . . . . . . . . . . . . . . . . . 84
4.5 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . .91
5 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93
5.1 Contributions . . . . . . . . . . . . . . . . . . . . . . . . . .93
5.2 Future Research Directions . . . . . . . . . . . . . . . . . . . 95
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97