Apple Inc.

Xiaoxing Li

Jianping Zhou

Rudolph Van Der Merwe

Jason Klivington

Sebastien X Beysserie

George E Williams

Damien J Thivent

Geoffrey T Anneheim

Anders M Holtsberg

Martin Lindberg

Paul M Hubel

Marius Tico

Marco Zuliani

Richard David Seely

Feng Li

Benjamin M Olson

Jia Xue

Libin Sun

Lech J Szumilas

Devices, methods, and non-transitory program storage devices are disclosed to provide enhanced images in multi-camera systems, e.g., by using information from images captured by cameras with different properties in terms of optics and/or sensors. In one embodiment, the techniques comprise: obtaining a first image from a first image capture device, wherein the first image has a first field of view (FOV) and a first set of quality characteristics; obtaining a second image from a second image capture device, wherein the second image has a second FOV and a second set of quality characteristics, and wherein the second FOV partially overlaps the first FOV; obtaining a neural network that produces a modified second image having a modified second set of quality characteristics determined by the neural network attempting to match the first set of quality characteristics; and generating an output image based, at least in part, on the modified second image.

Systems and methods for multi-sensor image enhancement

Foreign lighting effects, such as lens flare, are very common in natural images. In a two-camera-system, the two images may be fused together to generate one image of a better quality. However, there are frequently different foreign light patterns in the two images that form the image pair, e.g., due to the difference in lens design, sensor and position, etc. Directly fusing such pairs of images will result in non-photorealistic images, with composed foreign light patterns from both images from the image pair. This disclosure describes a general foreign light mitigation scheme to detect all kinds of foreign light region mismatches. The detected foreign light mismatch regions may be deemphasized or excluded in the fusion step, in order to create a fused image that keeps a natural-looking foreign light pattern that is close to what was seen by the user of an image capture device during an image capture preview mode.

Mismatched foreign light detection and mitigation in the image fusion of a two-camera system

Systems, methods, and computer readable media to rapidly identify and track an arbitrary sized object through a temporal sequence of frames is described. The object being tracked may initially be identified via a specified or otherwise known region-of-interest (ROI). A portion of that ROI can be used to generate an initial or reference histogram and luminosity measure, metrics that may be used to identify the ROI in a subsequent frame. For a frame subsequent to the initial or reference frame, a series of putative ROIs (each having its own location and size) may be identified and the 'best' of the identified ROIs selected. As used here, the term 'best' simply means that the more similar two frames' histograms and luminosity measures are, the better one is with respect to the other.

Fast histogram-based object tracking

Techniques to identify and track a pre-identified region-of-interest (ROI) through a temporal sequence of frames/images are described. In general, a down-sampled color gradient (edge map) of an arbitrary sized ROI from a prior frame may be used to generate a small template. This initial template may be used to identify a region of a new or current frame that may be overscan and used to create a current frame's edge map. By comparing the prior frame's template to the current frame's edge map, a cost value or image may be found and used to identify the current frame's ROI center. The size of the current frame's ROI may be found by varying the size of putative new ROIs and testing for their congruence with the prior frame's template. Subsequent ROI's for subsequent frames may be identified to, effectively, track an arbitrarily sized ROI through a sequence of video frames.

Fast template-based tracking

Techniques to permit a digital image capture device to stabilize a video stream in real-time (during video capture operations) are presented. In general, techniques are disclosed for stabilizing video images using an overscan region and a look-ahead technique enabled by buffering a number of video input frames before generating a first stabilized video output frame. (Capturing a larger image than is displayed creates a buffer of pixels around the edge of an image; overscan is the term given to this buffer of pixels.) More particularly, techniques are disclosed for buffering an initial number of input frames so that a 'current' frame can use motion data from both 'past' and &#x201c;future&#x201d; frames to adjust the strength of a stabilization metric value so as to keep the current frame within its overscan. This look-ahead and look-behind capability permits a smoother stabilizing regime with fewer abrupt adjustments.

Video image stabilization

Growing community of inventors

San Jose, CA, United States of America

Xiaoxing Li