A selfie system adapted to background environment using an omnidirectional camera

The widespread use of smartphones and social networking services has increased the popularity of selfie photos. Many users seek images that are visually appealing not only in terms of facial appearance but also in terms of important background objects and overall composition. However, capturing both the user’s face and major background landmarks within a single smartphone photograph is often difficult. This paper proposes a method that automatically generates well-composed selfie images using an omnidirectional camera. The proposed system detects the user’s face and identifies important background objects through face detection and semantic segmentation, then selects the optimal composition from multiple perspective-projection candidate images. We evaluated the proposed method using images captured at several tourist spots and compared the results with subjective evaluations.