US20260038195
2026-02-05
Physics
G06T15/205
An electronic device equipped with imaging sensors can capture a set of images to create an interactive selfie panorama. By using these images, a processing device generates a 3D model of the scene, which is then projected into an image space to produce a series of selfie images. This process involves capturing an initial burst of images, generating a map to identify subjects, and prompting the user to capture additional images to enhance the final output.
The device employs an iterative process to develop a differentiable 3D model of the scene. This model is refined by comparing measured and estimated images, adjusting model parameters to minimize discrepancies. The process includes creating a depth map from each image and classifying pixels into semantic categories, which helps in updating model parameters effectively.
The 3D model allows for the generation of a burst of selfie images from different perspectives. Perspective information is used to render each image accurately, and a machine learning model may be trained to predict these renderings. The final image is selected based on a metric comparison, and users can choose their preferred image from the rendered options.
The device provides prompts to guide users in capturing additional image bursts, ensuring comprehensive coverage of the scene. This interactive approach helps in creating undistorted and high-quality selfies, overcoming the limitations of traditional selfie cameras with restricted fields of view and potential subject distortion.
This technology is applicable in devices requiring advanced image processing capabilities, particularly for enhancing selfie quality. It leverages machine learning for model optimization and offers users an interactive experience to produce aesthetically pleasing images. The system's adaptability to various perspectives and conditions makes it a versatile tool in modern imaging applications.