Abstract: Techniques for removal of obstructions in a digital image scene are described, in which target and source digital images that exhibit parallax, one to another, are obtained that were captured together by an image capture device at a similar point in time using two different lenses of the image capture device. A foreground obstruction is identified based on displacement in apparent position of objects in the target and source digital images. The foreground obstruction is removed from the target digital image, such as by generating an obstruction mask that represents the location of the foreground obstruction and copying pixels from the source digital image to the target digital image based on the locations identified in the obstruction mask. The target digital image with the obstruction removed is output to a user interface or service provider system, for example.

Abstract: Methods, systems, and non-transitory computer readable storage media are disclosed for generating focused preview images that include the subjects of interest of digital images. For example, in one or more embodiments, the disclosed system utilizes a machine-learning model to generate a saliency map for a digital image to indicate one or more salient objects portrayed within the digital image. Additionally, in one or more embodiments, the system identifies a focus region based on focus information captured by a camera device at the time of capturing the digital image. Furthermore, the system can then utilize the saliency map and the focus region to generate a focused preview image. For instance, in one or more embodiments, the system crops the digital image based on an overlapping portion of the saliency map and the focus region to generate a focused preview image.

Abstract: Embodiments relate to automatic perspective and horizon correction. Generally, a camera captures an image as an image file. Capture-time orientation data from one or more sensors is used to determine the camera's attitude with respect to a defined reference frame. The orientation data and/or attitude can be registered into metadata of the image file and used to generate axis lines representative of the camera's reference frame. A reference line such as a horizon can be automatically identified from detected line segments in the image that align with one of the axis lines within a predetermined angular threshold. The reference line can be used to generate a camera transformation from a starting orientation reflected by the camera attitude to a transformed orientation that aligns the reference line with the reference frame. The transformation can be applied to the image to automatically correct perspective distortion and/or horizon tilt in the image.

Abstract: Methods, systems, and non-transitory computer readable storage media are disclosed for generating focused preview images that include the subjects of interest of digital images. For example, in one or more embodiments, the disclosed system utilizes a machine-learning model to generate a saliency map for a digital image to indicate one or more salient objects portrayed within the digital image. Additionally, in one or more embodiments, the system identifies a focus region based on focus information captured by a camera device at the time of capturing the digital image. Furthermore, the system can then utilize the saliency map and the focus region to generate a focused preview image. For instance, in one or more embodiments, the system crops the digital image based on an overlapping portion of the saliency map and the focus region to generate a focused preview image.

Abstract: Various embodiments receive frames as a stream captured during a camera session. During the camera session, faces and facial features of each face are detected from the frames. Then, each face in each frame is assigned a score based on the detected facial features. Using the scores, a candidate frame is selected for each individual face to represent a “best” representation of that face. In addition, an overall score is calculated for each frame based on a combination of assigned scores for the faces in the frame. Then, a reference frame is located from the frames based on the overall score for a respective frame. Faces from the candidate frames are then merged onto the reference frame, and an output image is generated for display.

Abstract: Techniques for removal of obstructions in a digital image scene are described, in which target and source digital images that exhibit parallax, one to another, are obtained that were captured together by an image capture device at a similar point in time using two different lenses of the image capture device. A foreground obstruction is identified based on displacement in apparent position of objects in the target and source digital images. The foreground obstruction is removed from the target digital image, such as by generating an obstruction mask that represents the location of the foreground obstruction and copying pixels from the source digital image to the target digital image based on the locations identified in the obstruction mask. The target digital image with the obstruction removed is output to a user interface or service provider system, for example.

Abstract: Various embodiments receive frames as a stream captured during a camera session. During the camera session, faces and facial features of each face are detected from the frames. Then, each face in each frame is assigned a score based on the detected facial features. Using the scores, a candidate frame is selected for each individual face to represent a “best” representation of that face. In addition, an overall score is calculated for each frame based on a combination of assigned scores for the faces in the frame. Then, a reference frame is located from the frames based on the overall score for a respective frame. Faces from the candidate frames are then merged onto the reference frame, and an output image is generated for display.

Abstract: A reflection removal system is capable of automatically removing from digital images reflections related to photographic illumination sources (“flash”), at the time of capture, without user intervention. In an embodiment, the reflection removal system receives a set of digital images taken at substantially the same time from a camera, where one of the image is affected by flash. The images are divided into blocks, and a threshold is determined for each block, indicating the brightest pixel value related to the content of the block. The reflection removal system compares each pixel in the flash-affected image to the corresponding threshold, and generates a digital mask based on the comparison. The reflection removal system creates a corrected image based on the digital mask, such that pixels affected by flash are modified based on corresponding pixels in the other images. In an embodiment, the corrected image is color blended.