Image stitching in the presence of a full field of view reference image

Abstract

Systems and methods for obtaining a seamless, high resolution, large field of view image comprise capturing a plurality of Tele images in a scene using a scanning Tele camera, each captured Tele image having an associated Tele field of view FOVT, retrieving a R image having a respective R image scene with a field of view greater than FOVT, analyzing the R image for defining an order of scanning positions according to which the folded Tele camera scans a scene to capture the plurality of Tele images, aligning the plurality of Tele images and the R image to obtain aligned Tele images, and composing the aligned Tele images into an output image. The output image may include at least parts of the R image and may be one of a stream of output images.

Description

FIELD

The subject matter disclosed herein relates in general to image stitching and in particular to methods for stitching images taken with a small FOV camera in the presence of a large FOV image.

BACKGROUND

Multi-aperture cameras (or multi-cameras) are the standard for high-end mobile handheld devices (“mobile devices”, e.g. smartphone, tablet, etc.). A multi-camera usually comprises a wide field of view (FOV_W) or “wide angle” camera (“Wide” camera or “W camera”), and one or more additional cameras, either with the same FOV, with a narrow FOV (Telephoto or “Tele” camera or “T” camera with Tele FOV_T) or an ultra-wide camera with a FOV_UW>FOV_W (“UW camera”).

The Tele-Wide or Tele-Ultra-Wide multi-camera (which in the following may be referred to as a “Tele-Reference” multi-camera) combines benefits of both cameras to overcome shortcomings. It captures a large reference camera FOV_R of the scene with the reference camera (“R camera”) that provides reference (R) images with a R camera resolution (“RES_R”) and R signal-to-noise ratio (SNR_R) and uses the higher Tele camera resolution (“RES_T”) and SNR (SNR_T) to zoom into the scene with FOV_T. However, as the Tele camera resolution increases, FOV_T narrows, so that only a fraction of a scene can be captured.

A recent development is a scanning Tele camera that can scan a scene with its regular (native) FOV_T, so that it effectively covers a scanning FOV_T larger than native FOV_T. A scanning Tele camera is described for example in co-owned U.S. Pat. No. 10,578,948.

There is need and it would be beneficial to have methods for using reference image (R image) data and Tele (T) image data to create a new image with a “new” field of view FOV_N that fulfills FOV_T<FOV_N≤FOV_R, wherein the image resolution of the new image RES_N>RES_R and/or wherein the SNR of the new image SNR_N>SNR_R.

SUMMARY

In various embodiments, there are provided systems and methods for using reference image data and Tele image data to create a “new” image with a “new” field-of-view FOV_N that fulfills FOV_T<FOV_N≤FOV_R. The new image is a seamless, high resolution, large FOV image. The new image has a resolution RES_N greater than RES_R and/or a signal-to-noise ratio SNR_N greater than SNR_R. The new image can be a “super image” (“SI”), obtained by capturing and stitching two or more T images, or it can be a “super-wide image” (“SW”), obtained by capturing and using one or more T images to improve a R image or segments thereof with a super-resolution algorithm.

FIG. 1 illustrates one difference between (a) a known panorama image (or simply “panorama”) and (b) a super image disclosed herein. The panorama image has an inherent perspective distortion due to the lack of “ground truth” when correcting points-of-view (POVs) of the different images forming it, while a super image disclosed herein is not distorted.

A SI comprises at least two high resolution Tele images and a R image (with RES_R<RES_T) with a large FOV (e.g. FOV_W or FOV_UW) of the same scene, see FIG. 1(b). That is, a SI uses information regarding the same scene from a larger FOV reference image. In contrast, for creating a panorama as known in the art, in general no reference image is available. In a SI, each captured Tele image is processed and matched to its corresponding location in the larger FOV reference image, thus removing the normal panoramic image perspective distortion. In addition, each Tele image may be checked for capture issues such as focus miss, large amounts of noise, temporary occlusions and more, and, if necessary, is retaken at a later time.

FIG. 2 illustrates another difference between (a) a R image and (c) a SI obtained as disclosed herein. FIG. 2(b) shows the single Tele images used to compose the SI. Since a Tele camera with a large effective focal length (EFL) is used to capture the images, one achieves in the SI (c) a “natural” Bokeh effect. Natural Bokeh is defined as the amount of blurring present in the image data of FOV segments that are not in-focus.

FIG. 3 illustrates yet another difference between a known panorama and a SI obtained as disclosed herein. One can see a scenario where two Tele images, Tele1 and Tele2, have no information in an overlapping FOV region, but the images are not entirely without information. In this situation, a known panorama will fail to align the images while the SI will be able to align them using a R image.

Other differences between a known panorama and a SI include the ability in the SI acquisition to scan automatically, thereby enabling to determine the scanning position and order of the Tele camera in an educated fashion; the ability to detect and correct bad images; and increased robustness due to higher overlap with the ground truth instead of overlap between different Tele images, which requires a smaller amount of images to cover a desired FOV, since the overlap size demand between the Tele images is reduced.

In various embodiments, there is provided a method, comprising: providing a folded Tele camera configured to scan and capture a plurality of Tele images, each captured image having a RES_T, a SNR_T and a FOV_T; obtaining and analyzing a R image with FOV_R>FOV_T and with an image resolution RES_R<RES_T, and/or a R image with SNR_R<SNR_T; determining an order of one or more scanning FOV_T positions for consecutive captures of the Tele images; capturing a Tele image at each respective scanning FOV_T position; aligning the captured Tele images with segments of the R image to obtain aligned Tele images that are aligned with the R image; and using the aligned Tele images and the R image to create a new image having a field of view FOV_N≤FOV_R, wherein the image resolution of the new image RES_N>RES_R and/or wherein the SNR of the new image SNR_N>SNR_R.

In some embodiments, the R image is a Wide image having a FOV_W>FOV_T, the Wide image captured by a Wide camera included a multi-camera together with the folded Tele camera.

In some embodiments, the R image is an Ultra-Wide image having FOV_UW>FOV_W>FOV_T, the Ultra-Wide image captured by an Ultra-Wide camera included in a multi-camera together with the folded Tele camera.

In some embodiments, a method further comprises aligning each Tele image with the R image immediately after its capture and prior to the capture of an immediately following Tele image, analyzing each Tele image for faults, and if faults are detected in the Tele image, re-capturing the Tele image at a same FOV_T position, or, if faults are not detected in the Tele image, proceeding to capture an immediately following Tele image at a respective FOV_T position.

In some embodiments, a method further comprises analyzing the aligned Tele images for faults, and if faults are detected in a particular Tele image, re-capturing the particular Tele image at a same FOV_T position, or, if faults are not detected, using the aligned Tele images and the R image to create the new image.

In some embodiments, the folded Tele camera captures two or more Tele images at two or more respective FOV_T positions within FOV_R, and the aligned Tele images are composed to create a super image.

In some embodiments, the aligned Tele images and the R image are fed into an algorithm to create a super wide (SW) image having a field of view FOV_SW, wherein a FOV segment within FOV_R included in at least one FOV_T of the captured Tele images has a field-of-view union-FOV_T and wherein union-FOV_T<FOV_SW≤FOV_R.

In some embodiments, the folded Tele camera is a multi-zoom Tele camera having different zoom states for capturing Tele images having different respective zoom factors.

In some embodiments, the obtaining of the R image includes obtaining the R image from the Internet, from a cloud database, or from an Internet of Things device.

In some embodiments, a video stream formed by a sequence of a plurality of new images is output instead of single new image.

In some embodiments, a user or an algorithm selects a size of FOV_N and a position of FOV_N within a scene included in FOV_R.

In some embodiments, the analyzing of the R image includes using a saliency map of the R image for automatically selecting the scanning FOV_T positions and/or automatically selecting the FOV_N.

In some embodiments, the aligning of the captured Tele images to obtain a plurality of aligned Tele images includes localizing the T image data with respect to the R image data.

In some embodiments, the Tele camera has an effective focal length of 7-10 mm, of 10-20 mm, or of 20-40 mm.

In some embodiments, the determining an order of one or more scanning FOV_T positions is performed so that each of the one and more Tele images exhibits a specific amount of natural Bokeh.

In some embodiments, the determining an order of one or more scanning FOV_T positions is performed so that a composed new image covers a maximal FOV according to a mechanical limitation of the scanning.

In some embodiments, the determining an order of one or more scanning FOV_T positions is performed so that the new image covers a region of interest selected by a user or by an algorithm.

In some embodiments, the determining an order of one or more scanning FOV_T positions is performed so that each T images include scene segments having a specific depth range or include scene segments that do not exceed a specific depth threshold.

In some embodiments, the determining an order of one or more scanning FOV_T positions is performed so that moving objects are captured first, and after the moving objects are captured, stationary objects are captured.

In some embodiments, the determining an order of one or more scanning FOV_T positions is performed so that a desired coverage of FOV_R with a plurality of FOV_T is performed in a fastest manner.

In some embodiments, the determining an order of one or more scanning FOV_T positions includes determining an order to follow a moving object with an object tracker.

In some embodiments, the determining an order of one or more scanning FOV_T positions includes capturing an object in a Tele image with specific FOV_T to improve RES or SNR of a similar object included in FOV_N but not included in the specific FOV_T.

In some embodiments, the determining an order of one or more scanning FOV_T positions includes capturing a moving object with more than one FOV_T at two significantly different points in time.

In some embodiments, FOV_N>FOV_T.

In some embodiments, the Tele images are additionally aligned with respect to each other to obtain aligned Tele images that are aligned with the R image and with each other.

In some embodiments, the R image includes a plurality of Wide images.

In some embodiments, the scanning Tele camera covers a scanning range larger than 25% of FOV_W and/or FOV_UW or larger than 50% of FOV_W and/or FOV_UW.

In some embodiments, the scanning Tele camera resolution RES_T>2×RES_W and/or SNR_T>2×SNR_W, or RES_T>4×RES_W and/or SNR_T>4×SNR_W.

In some embodiments, the faults are selected from the group consisting of motion blur, electronic noise, rolling shutter, defocus blur and incorrect image alignment or obstructions. In some embodiments, the faults are mechanical faults.

In some embodiments, the composing the aligned Tele images into the super image includes composing the aligned Tele images together with the R image into the super image.

In some embodiments, the determining an order of two or more FOV_T positions is performed so that capturing a minimal number of T images is required.

In some embodiments, the determining an order of two or more FOV_T positions is performed so that Tele images including specific scene characteristics within their respective FOV_T s may be captured consecutively, and wherein the scene characteristics may be visual data such as texture or physical data such as brightness, depth or spectroscopic composition of a scene.

In some embodiments, the determining an order of two or more scanning FOV_T positions is performed so that a moving object is removed from a scene included in FOV_R.

In some embodiments, the determining an order of two or more scanning FOV_T positions is performed to create a duplication of a moving object in a scene included in FOV_R.

In some embodiments, the determining an order of two or more scanning FOV_T positions is performed so that each Tele image overlaps with another Tele image.

In some embodiments, the determining an order of two or more scanning FOV_T positions is performed so that one or more objects of interest are located in a center region of a FOV_T and not in an overlap region.

In some embodiments, the multi-zoom Tele camera is a continuous-zoom Tele camera. In some embodiments, the multi-zoom Tele camera is a dual-zoom-state Tele camera.

In some embodiments, the R image is a Tele image having a first ZF (ZF1) and the Tele images that are captured consecutively according to the scanning order have a second zoom factor (ZF2), wherein ZF1≤1.25×ZF2.

A method disclosed herein is not limited to a specific multi-camera module, and could be used for any combination of cameras as long as the combination consists of at least one camera with some scanning capabilities, not limited to 2D scanning.

BRIEF DESCRIPTION OF THE DRAWINGS

Non-limiting examples of embodiments disclosed herein are described below with reference to figures attached hereto that are listed following this paragraph. Identical structures, elements or parts that appear in more than one figure are generally labeled with a same numeral in all the figures in which they appear. If identical elements are shown but numbered in only one figure, it is assumed that they have the same number in all figures in which they appear. The drawings and descriptions are meant to illuminate and clarify embodiments disclosed herein and should not be considered limiting in any way. In the drawings:

FIG. 1 illustrates a key difference between (a) a regular panorama image and (b) a super image;

FIG. 2 illustrates another key difference between a regular panorama image and a super image obtained as disclosed herein;

FIG. 3 illustrates yet another key difference between a regular panorama image and a super image obtained as disclosed herein;

FIG. 4A shows a flow chart of method embodiments for obtaining a new image disclosed herein;

FIG. 4B shows a flow chart of SWM embodiments for obtaining a new image, wherein the new image is a SW disclosed herein;

FIG. 5 illustrates the FOVs of dual-camera output images;

FIG. 6 illustrates an embodiment of a R image retrieval process;

FIG. 7 illustrates another embodiment of a R image retrieval process;

FIG. 8A illustrates an embodiment of Tele scanning order determination for creating a SI or SW;

FIG. 8B illustrates another embodiment of determining a Tele scanning order for creating a SI or SW;

FIG. 9 illustrates determination of Tele FOV positions according to certain criteria;

FIG. 10A illustrates Tele camera scanning order in an embodiment of Tele scanning order determination based on the detection of a moving object in a scene;

FIG. 10B illustrates a super image resulting for the scanning in FIG. 10A;

FIG. 11A illustrates Tele camera scanning order in another embodiment of Tele scanning order determination based on the detection of a moving object in a scene;

FIG. 11B illustrates a super image resulting for the scanning in FIG. 11A;

FIG. 12A illustrates Tele camera scanning order in yet another embodiment of Tele scanning order determination based on the detection of a moving object in a scene;

FIG. 12B illustrates a super image resulting for the scanning in FIG. 12A;

FIG. 12C shows a first example of a T scanning order that prioritizes capturing moving objects over stationary objects;

FIG. 12D shows a SW resulting from the scanning order in FIG. 12C;

FIG. 13 is a flow chart showing details of a typical setup of the image alignment step;

FIG. 14 illustrates a few cases in which the fault detection step will raise an error;

FIG. 15A is a flow chart showing steps of a method for SI composition;

FIG. 15B is a flow chart showing steps of a SWM;

FIG. 16A shows a super image without the use of any color correction step;

FIG. 16B shows a super image having the color correction step;

FIG. 17 shows schematically an embodiment of a mobile device that includes multi-aperture cameras with at least one scanning Tele camera.

DETAILED DESCRIPTION

Returning now to the figures, FIG. 4A shows a flow chart of method embodiments for obtaining a new image with RES_N>RES_R and/or SNR_N>SNR_R disclosed herein. A first method may be referred to as Super-Image method (or “SIM”) for creating a super-image as described herein (for SI composition see FIG. 15A). A second method may be referred to as Super-Wide image method (or “SWM”) for creating a super-wide image as described herein (for SW composition see FIG. 15B). An algorithm performing the methods may run on a processor such as Application processor (AP) 1740 operatively coupled to a first (R) camera and a second (Tele) camera, as shown in FIG. 17. In some embodiments for obtaining a SI, an initial Tele image at an initial scanning position is captured in step 402. A R image is retrieved in step 404. In some embodiments, the R image is a W image or an UW image taken with respective cameras. In some embodiments, the R image is taken at a significantly different time than the Tele images captured in step 402 and/or 408, e.g. few seconds, hours or even years before the capturing of the Tele images. In other embodiments, the R image is retrieved from an external database using the initial Tele image as a query, see below. A next (also referred to as “subsequent”, “sequential” or “consecutive”) scanning position of the Tele camera, determining a respective FOV_T, is selected or updated after the initial and each subsequent Tele image capture in step 406.

In step 408, a subsequent Tele image is acquired (captured) using the scanning position selected or updated in step 406. For a SIM, the subsequently acquired Tele image is aligned with previously found Tele images that have some shared FOV and with the R image in step 410 to obtain an aligned Tele image. For a SWM, the subsequently acquired Tele image is aligned with the R image in step 410 to obtain an aligned Tele image. The aligned Tele image is analyzed for faults in step 412 and, based on the detected faults, a subsequent scanning position is updated by returning to step 406. Steps 406-412 are repeated until the desired coverage of the R image has been achieved. Afterwards, the SI or SW are composed as described in FIG. 15A for SIM and in FIG. 15B for SWM in step 414 to form one cohesive new image with FOV_N>FOV_T and RES_N>RES_T and/or SNR_N>SNR_T, which can be a SI or a SW.

In some embodiments, image composition step 414 may be performed after all the Tele images are acquired and aligned as described above. In other embodiments, image composition step 414 may be performed after each iteration of Tele image acquisition and image alignment steps 406-412, to perform “on the fly” blending with intermediate viable results. In such embodiments, a SI exists after each iteration of steps 406-412.

FIG. 4B shows a flow chart of SWM embodiments for obtaining a new image with RES_N>RES_R and/or SNR_N>SNR_R wherein the new image is a SW disclosed herein. An algorithm performing the methods may run on AP 1740 operatively coupled to a first (R) camera and a second (Tele) camera, as shown in FIG. 17. In step 422, a R image is retrieved. In step 424, features present in the R image are extracted as known in the art and clustered in 426. The features may be clustered by their distance from the mobile device 1700, or by their location within the R image's FOV or otherwise, e.g. by their similarity in a semantic sense, e.g. by semi-image segmentation such as “SLIC Superpixels” described by Radhakrishna Achanta, Appu Shaji, Kevin Smith, Aure-lien Lucchi, Pascal Fua, and Sabine Susstrunk, SLIC Superpixels, EPFL TechnicalReport 149300, June 2010. Based on the result of the clustering, in step 428 a T scanning order is determined. Optionally and in case that T camera 1710 is a multi-zoom T camera having different zoom states, additionally a suitable zoom factor (ZF) may be determined. Examples for a T scanning order are given in FIG. 12C-D. In step 430, T images are captured according to the T scanning order determined in step 428.

Steps 432-440 describe the process of aligning the T images captured in step 430 with the R image retrieved in step 422. Further details on the image alignment are described in FIG. 13. As known in the art, T images captured with a scanning T camera based on 1 OPFE undergo a “Roll effect”, so that they are rectified in step 432. Features in the T images are extracted in step 434. For each T image, the features extracted in step 434 are matched with the features in R image that were extracted in step 424. Steps 438-440 describe the fine-tuning of the alignment process. In step 438, in case the rectification performed in step 432 was found to be insufficient in step 436, the T image is additionally adjusted in terms of warping for reducing Roll effect. In step 440, a localization of the T image data with respect to the R image data is performed. For example, a cross correlation calculation as known in the art may be performed for localizing the T image data within the R image. As an example and typically, before the localization, image points of a same object point may e.g. deviate by more than 25 pixels or by more than 50 pixels or by more than 100 pixels between the R and T images. We assume a pixel size of about 11 μm. After the localization, image points of a same object point may e.g. deviate by less than 20 pixels or by less than 10 pixels or even by less than 2 pixels between the R and T images. For SWM, it may not be required to align the T images with respect to each other, but each T image may be aligned with the R image only.

In step 442, the R image and the aligned T images are fed into a super-resolution algorithm. Relevant super-resolution algorithms are described for example in Daniel Glasner et al., “Super-Resolution from a Single Image”, ICCV, 2009, Tamar Rott Shaham et al., “SinGAN: Learning a Generative Model from a Single Natural Image”, ICCV, 2019, arXiv:1905.01164, or Assaf Shocher et al., “Zero-Shot Super-Resolution using Deep Internal Learning”, 2017, arXiv:1712.06087.

A new image having RES_N>RES_R and/or SNR_N>SNR_R is output in step 444. In general, FOV_N is larger than the union of all FOV_T s that are fed into the super-resolution algorithm in step 442, i.e. FOV_N>union-FOV_T. Union-FOV_T represents the FOV within FOV_R which is included in at least one FOV_T of one of the T images captured in step 428.

The FOV_T scanning may be performed by actuating (e.g. for rotation) one or more optical path folding elements (OPFEs) of the scanning Tele camera. Fast actuation may be desired. Actuation may be performed in 2-20 ms for scanning e.g. 2°-5° and in 10-70 ms for scanning 15-25°. A scanning Tele camera may have a maximal diagonal scanning range of 60°. “Maximal diagonal scanning range” is defined by the center of the FOV in the maximum state bottom-left of a center FOV and the center of the FOV in the maximum state top-right of a center FOV. For example and referring to FOV diagonal, a scanning T camera having FOV_T=20° and 60° scanning range covers an overall FOV of 80°. A diagonal scanning range of 40° may cover around 60-100% of a FOV_W. The scanning Tele camera may have an of EFL=7 mm-40 mm. Typical zoom factors (ZF) may be 2×-10× zoom with respect to a W camera hosted in the same mobile device, meaning that an image of a same object captured at a same distance is projected at a size 2×-10× larger on the image sensor of the T camera than on the W camera. Assuming that a same sensor is used in R camera and T camera, the image resolution scales linearly with the ZF. For same sensors, typically, RES_T>2× RES_W. In some examples, RES_T>5× RES_W.

FIG. 5 illustrates exemplary FOVs of a dual-camera image, for a dual-camera that comprises for example a W camera (acting as a R camera) and a scanning T camera. Tele FOV 0 shows one exemplary T scanning position superposed on the Wide FOV and FOV 1 shows another exemplary T scanning position superposed on the Wide FOV. Such a setup enables the use of the W image as the R image in step 404.

FIG. 6 illustrates an embodiment of a R image retrieval process. Three T images Tele1, Tele2 and Tele3 are acquired in steps 402 and/or 408. These images are then used to search an external database (e.g. using Google) for a similar image, which is then retrieved and used as the R image.

FIG. 7 illustrates another embodiment of a R image retrieval process. Here, the retrieval is performed using GPS location from a mobile device. The figure shows the current mobile device's GPS location and two T images (Tele1 and Tele2) captured in step 402 and/or 408, and “GPS” R images retrieved from an external database using the GPS location as a query.

FIG. 8A illustrates an embodiment of Tele (T) scanning order determination for creating a SI or SW (step 406 or step 428 respectively). The scanning order is indicated by numbering the images from 1-9, i.e. T image 1 having Tele FOV 1 is the first T image captured, T image 2 having Tele FOV 2 is the second T image captured, etc. Here, the scanning positions are determined based on the maximal coverage of the T camera's scan range. The figure shows the maximal possible scanning range coverage of the T camera superimposed on a Wide FOV, and T images with Tele FOVs from FOV1 to FOV9 selected so as to achieve the maximal coverage. For creating a SW, the overlap between the single Tele images 1-9 is optional.

FIG. 8B illustrates another embodiment of determining a T scanning order for creating a SI or SW (step 406 or step 428 respectively). The scanning order is indicated by numbering as see FIG. 8A. Here, the scanning positions are determined based on the maximal coverage of the scan range and additionally on capturing this maximal coverage in the shortest overall capture time. The capture time includes consecutively moving FOV_T to the target scanning position and capturing one or more T images at this position. In this embodiment, by minimizing the FOV_T movement between T image 3 having Tele FOV 3 and T image 4, and between T image 6 and T image 7, a shorter overall capture time compared to the scanning order shown in FIG. 8A is achieved.

It is noted that determining a scanning order includes determining the respective FOV_T position, meaning that FOV_T positions and their scanning order are determined.

FIG. 9 illustrates determining Tele FOV positions for creating a SI and or a SW according to the criteria of maximal coverage of a user's region of interest (ROI) choice. The figure shows the user's requested ROI as a dashed rectangle on the Wide (as reference) image FOV and the T images determined so as to achieve the maximal coverage of this ROI choice.

In other embodiments for SIM and SWM, the scanning positions may be determined based on the maximal coverage of an object of interest or ROI as obtained from an algorithm, e.g. from a Saliency map, for example as described in “Salient Object Detection: A Discriminative Regional Feature Integration Approach” by Jiang et al. or as in “You Only Look Once: Unified, Real-Time Object Detection” by Redmon et al. The FOV of a SI or a SW may be selected based on the Saliency map.

In yet other embodiments for SIM, the scanning positions may be determined such that specific features within an ROI are located in a center region of a FOV_T and not in an overlap region. A specific feature may be for example the face of a person. Locating specific features in a center region may avoid stitching artifacts in the SI's FOV segments where the ROI is located, e.g. by applying “stitching seams” in the FOV covered by the specific feature.

In yet other embodiments for SIM and SWM, scanning positions may be determined so that a minimal number of T image captures is required for a given selected ROI covering a particular FOV which is larger than FOV_T, e.g. for reducing power consumption and capture time.

In yet other embodiments for SIM and SWM, a criterion for determining an order of scanning position may be based on artistic or visual effects such as e.g. a desired amount of natural Bokeh. The amount of natural Bokeh depends on differences in the object-lens distance of foreground objects (in-focus) and background objects (out-of-focus). A scanning position criterion may e.g. be an image background with uniform natural Bokeh.

In yet other embodiments for SIM and SWM, a criterion for determining an order of scanning position may be based on desired data for computational photography. Such data may be for example stereo image data including T image data and image data from the R image. From stereo image data of a single FOV_T and the overlapping image FOV segment of the FOV_R, a stereo depth map covering FOV_T may be calculated as known in the art, e.g. by triangulation. The stereo depth map may enable application of artificial Bokeh algorithms to the R image or to the SI. In some embodiments, the SI output in step 414 may not be an image including visual data, but an output that includes stereo depth data.

In other embodiments, a scanning order criterion may include desired artistic SI effects. Such effects may be created by synchronizing T image capture and FOV scanning, wherein capture happens during FOV movement, so that a motion blur effect in the T image is achieved. For this, a scanning order criterion may be a desired amount of motion blur of a specific scene segment.

In yet other embodiments for SIM and SWM, a criterion for scanning position determination may be based on a depth estimation of the scene included in the R image. For example, one may select scanning positions so that single T images include scene segments having a specific depth range (i.e. a specific camera-object distance range) or include scene segments that do not exceed a specific depth threshold. In another example, one may select scanning positions so that single T images include ROIs covering a particular FOV size. As an example, a scanning order criterion may be to capture scene segments having similar depths or including ROIs of particular FOV sizes consecutively. This may be beneficial for a scanning camera that may have not one fixed FOV (i.e. zoom state) but different FOVs (zoom states). For fast SI or SW capture, one may prefer to capture FOV segments with identical zoom states consecutively (sequentially), as it may e.g. be desired to minimize number of (time-consuming) zoom state switches. As another example, a scanning order criterion may be to capture scene segments having similar depths consecutively, because this may minimize the amount of time required for re-focusing the T camera between single T image captures and may also facilitate the alignment of the T images.

In yet another embodiment for SIM and SWM, a scanning order criterion may be that T images comprising specific scene characteristics within their respective FOV_T s may be captured consecutively. In some embodiments, T images with similar scene characteristics within their respective FOV_T s may be captured consecutively. Scene characteristics may be visual data such as texture. Scene characteristics may be physical data such as brightness, depth or spectroscopic composition of a scene. A spectroscopic composition may be defined by the intensity values of all wavelengths present in the scene.

FIGS. 10A, 10B, 11A, 11B, 12A and 12B illustrate three examples for criteria for determining a T scanning order for creating a SI based on the information related to a moving object (a runner, the arrow in FIG. 11A indicating the direction of her movement) in the scene. Views A in each figure show the T scanning order in a flow chart (arrows 1, 2, 3). View B in each figure shows the resulting SI. There are several options for determining a T scanning order, as follows.

One can determine the order of capturing the T images such that the moving object will not appear in the scene at all, as illustrated in FIG. 10A. Alternatively, one can determine the T scanning order such that the moving object will appear in the scene only once, as illustrated in FIG. 11A. Further alternatively, one can determine the T scanning order to create duplication of the moving object in the scene as illustrated in FIG. 12A.

The T scanning order (i.e. the scanning order criteria) may alternatively be based on camera or scene properties. In some embodiments, a scanning order criterion may be based on fast SI capture. In some embodiments, the SI output in step 414 or the SW output in step 444 may not be an image including visual data, but it may be an output including spectroscopic data, stereo depth data or other image data that is generated by computational photography or physical analysis.

In some embodiments, a plurality of sub-SIs that form a single SI may be captured in the FOV of a R image simultaneously, i.e. in a single capture process as described in FIG. 4. A sub-SI is defined by being a part of a single SI, wherein the single SI covers a larger FOV of a scene than the sub-SI. A sub-SI has no FOV overlap with respect to all other sub-SIs. A scanning order criterion for capturing the single SI may be fast SI capture. Fast SI capture may be achieved by minimizing delay time between the capture of single T images caused by OPFE actuation. In other embodiments, a scanning order criterion for capturing the single SI may be fast capture of the specific sub-SI that comprises the fastest moving object in the scene of the single SI.

FIGS. 12C-12D show the same scene shown in FIGS. 10A, 11A and 12A. FIG. 12C shows the T scanning order in a flow chart. FIG. 12D shows the resulting SW.

In contrast with SIM, in a SWM for increasing RES or SNR in a segment of FOV_R one must not necessarily capture a T image having a FOV_T that includes this very FOV_R segment. It may be sufficient to capture a T image that includes similar features present in the same scene. As an example and with reference to FIG. 12C, for increasing RES or SNR in a segment of FOV_R that includes bench 2 (marked “2” in FIG. 12C), it may be sufficient to use a T image that includes bench 1 (marked “1” in FIG. 12C), as bench 1 is an object that is similar to bench 2. Similarly, one may use a T image including a segment of the meadow appearing in Tele FOV 3 to increase the resolution of the new image at a location that is e.g. included in Tele FOV4, but not in Tele FOV 3.

Furthermore, for SWM the T images must not necessarily be aligned with each other, but only with the R image. Therefore, the captured T images must not necessarily include an overlapping FOV, which is required for SIM.

There are several options for determining a T scanning order, as follows.

FIG. 12C shows a first example of a T scanning order that prioritizes capturing moving objects (such as the runner) over stationary objects (such as the city landscape). In other examples, a first Tele FOV1 may be capture such as shown here, but a second Tele FOV2 that may still include the legs of the runner may be captured at a significantly later time, e.g. when the runner moved already to another position. A “significantly later time” may be for example a time that is 100 ms later or 250 ms later or 500 ms later or even later than that, e.g. is later. In this example, one may follow an object with an object tracker for capturing a T image.

In another example, a T scanning order is determined so that a desired coverage of FOV_R with a plurality of FOV_T is performed in a fastest manner.

In yet another example and for a Tele camera which is a multi-zoom camera, a T scanning order is determined so that a desired coverage of FOV_R with a desired zoom factor (ZF) is performed in a fastest manner. A user or an algorithm may select the desired ZF. One criterion for selecting the ZF may be a desired ratio of RES_T/RES_R and/or of SNR_T/SNR_R, another criterion may be a desired FOV_T. In some embodiments, the R image may be a Tele image which is captured with a first ZF (ZF1) and the Tele images that are captured consecutively according to the order have a second ZF (ZF2), wherein ZF1<ZF2, for example ZF1≤1.1×ZF2, ZF1≤1.25×ZF2, ZF1≤2×ZF2.

In yet another example and for a Tele camera which is a multi-zoom camera, a T scanning order is determined so that Tele images with a same ZF are captured consecutively. For example, first all Tele images with a particular first ZF (ZF1) are captured, and afterwards all Tele images with a particular second ZF (ZF2) are captured.

FIG. 13 is a flow chart showing details of a typical setup of image alignment step 410 for creating a SI or steps 432-440 for creating a SW. The inputs to these steps are the newly captured T image, the previously found features from all the T images captured thus far, and the R image (sub-steps 1302-1304). A feature extraction sub-step 1306 is followed closely by a feature matching sub-step 1308. These procedures can be such as described in, but not limited to, “ORB: an efficient alternative to SIFT or SURF” by Rublee et al., International Conference on Computer Science, 2011, or as in any other feature scheme. Next is a homography estimation sub-step 1310, such as described in, but not limited to, “Computing homography with RANSAC algorithm: A novel method of registration” by Li et al., Proceedings of SPIE—The International Society for Optical Engineering, January 2004, or as in any other homography estimation algorithm. In some embodiments, not a homography estimation may be used, but some other geometric transformation, e.g. an affine transformation or a non-affine transformation. A sub-step 1312 performs image warping according to the homography found in sub-step 1310 to align the resulting T image. In some embodiments, the image alignment is done at first between T images, while using the R image after the alignment for fine-tuning. In other embodiments, the image alignment may be done with each T image aligned to the R image while using the different T image matchings for fine-tuning, or by other combinations of image alignments between the different images. In some embodiments, a localization step such as described in step 440 may be performed.

FIG. 14 illustrates a few cases in which fault detection step 412 will raise an error. The cases include, but are not limited to, (a) motion blur, (b) electronic noise, (c) defocus blur, (d) rolling shutter artifacts, (e) incorrect image alignment (failure of the image alignment in step 410 for any reason) and (f) obstructions in the scene. An obstruction occurs if a first object covers segments of the FOV of a second object when the user desires to capture the second object. A R image may be used to detect motion blur or incorrect image alignment or obstructions in the scene, a T image analysis may be performed for detecting electronic noise. An error in fault detection step 412 may also be raised for other reasons one would generally, e.g. in the context of single image, not refer to as a “fault”, but which may still cause an expected decrease in the quality of the SI or SW composed in step 414. Such reasons raising an error may be related to the camera specifications used for capturing the T image. Some of these specifications may e.g. be related to exposure, camera lens modes, or image sensor modes.

Some reasons may be related to scene characteristics that were not identified in the R image analysis. Consider for example a bright oscillating light source in FOV_N. The light source may have been “Off” when the R image was captured, but it may have been “On” when the respective T image was captured, causing large differences in the T camera parameters deployed for this T image in contrast to prior or consecutive T images. In such a scenario re-capturing the T image with the light source “Off” may be desired.

An additional fault reason may relate to mechanical faults, e.g. the OPFE did not reach the desired location accurately, and therefore issues in the alignment of the image may occur and the image needs to be recaptured.

FIG. 15A is a flow chart showing steps of a method for SI composition, i.e. details of a typical setup of the image composition step 414. The inputs are the two or more aligned T images and the previously found features from all the T images and from the R image (sub-steps 1502-1504). The first algorithmic sub-step is color correction sub-step 1506. Sub-step 1506 uses color correction based on the R image and not the other T images, and is therefore superior to other related color correction procedures used in panorama images as known in the art, which only use information from the image overlap to perform color correction. The color correction sub-step itself may be performed, for example, as in “Color Correction for Mobile Panorama Imaging” by Xiong et al., ICIMCS '09: Proceedings of the First International Conference on Internet Multimedia Computing and Service, 2009, pp. 219-226.

The influence of color correction step 1506 on the SI is shown in FIGS. 16A and 16B. FIG. 16A shows a SI without the use of any color correction step. The images showing different FOV_T s exhibit significant color differences between each other. T color differences are unnatural, i.e. they do not stem from differences in the scene. FIG. 16B shows a SI including the color correction step. The color correction ensures a seamless output image. Seam estimation sub-step 1508 finds the optimal seam along which one wants to combine two adjacent T images in the overlap region. This step can use seams such as, but not limited to, “Coarse-to-fine Seam Estimation for Image Stitching” by Liao et al. Image blending step 1510 may be done as known in the art, for example according to “A multiresolution spline with application to image mosaics” by Burt et al. ACM Transactions on Graphics, Vol. 2. No. 4, October 1983, pages 217-236.

FIG. 15B is a flow chart showing steps of a SWM i.e. details of a typical setup of the steps 442 and 444. In step 1522, one or more T images that are each aligned with the R image and a R image are fed into super-resolution algorithm 1526. A new image is output in step 1528.

FIG. 17 shows schematically an embodiment of a mobile device numbered 1700 and including multi-cameras with at least one scanning Tele camera. Mobile device 1700 comprises a first scanning T camera module 1710 that includes an OPFE 1712 for FOV scanning, and a first lens module 1714 that forms a first image recorded by a first image sensor 1716. In other embodiments, camera module 1710 may include two OPFEs for FOV scanning. A first lens actuator 1718 may move lens module 1714 for focusing and/or optical image stabilization (OIS). An OPFE actuator 1720 may move the OPFE for (among others) scanning. In some embodiments, camera module 1710 may be a scanning Tele multi-zoom camera that can switch between Tele lens states having different zoom factors (ZF). With changing ZF, FOV_T changes. A Tele multi-zoom camera with two discrete ZFs is described for example in co-owned international patent application PCT/IB2020/051405.

Mobile device 1700 may further comprise a R (e.g. W or UW) camera module 1730 with a FOV larger than the FOV of camera module 1710. Camera module 1730 includes a second lens module 1732 that forms an image recorded by a second image sensor 1734. A second lens actuator 1736 may move lens module 1732 for focusing and/or OIS.

In some embodiments, first calibration data may be stored in a first memory 1722 of a camera module, e.g. in an EEPROM (electrically erasable programmable read only memory). In other embodiments, first calibration data may be stored in a third memory 1750 such as a NVM (non-volatile memory) of mobile device 1700. The first calibration data may comprise calibration data for calibration between sensors of R camera module 1730 and of T camera module 1710. In some embodiments, second calibration data may be stored in a second memory 1738. In some embodiments, the second calibration data may be stored in third memory 1750. The second calibration data may comprise calibration data between sensors of R camera module 1730 and T camera module 1710.

Mobile device 1700 may further comprise an application processor (AP) 1740. In use, AP 1740 may receive respective first and second (reference) image data from camera modules 1710 and 1730 and supply camera control signals to camera modules 1710 and 1730. In some embodiments, AP 1740 may receive first image data from camera module 1710 and R image data from third memory 1750. In other embodiments, AP 1740 may receive calibration data stored in a first memory located on camera module 1710 and in a second memory located in camera module 1730. In yet another embodiment, AP 1740 may receive R image data stored in third memory 1750. In yet another embodiment, AP 1740 may retrieve R images from an external database. AP 1740 includes an image analyzer 1742 for analyzing R images (e.g. for scene understanding and defining a Tele scanning order) and T images (e.g. for fault detection), a FOV scanner 1744 that calculates an OPFE control signal (e.g. for implementing a Tele scanning order) and an image generator 1744 for composing new images as outlined in steps 402-414 and in steps 1502-1510 (for SIM) and in steps 422-444 and in steps 1522-1528 (for SWM).

While this disclosure has been described in terms of certain embodiments and generally associated methods, alterations and permutations of the embodiments and methods will be apparent to those skilled in the art. The disclosure is to be understood as not limited by the specific embodiments described herein.

All references mentioned in this application are incorporated herein by reference in their entirety. It is emphasized that citation or identification of any reference in this application shall not be construed as an admission that such a reference is available or admitted as prior art.

Claims

1. A method, comprising: providing a folded Tele camera configured to scan and capture a plurality of Tele images, each captured image having a Tele image resolution (REST), a Tele image signal-to-noise-ratio (SNRT) and a Tele field of view (FOVT);obtaining and analyzing a reference (R) image with a R field of view FOVR>FOVT having a R image resolution RESR<REST, and/or a R image with a signal-to-noise-ratio SNRR<SNRT;determining an order of one or more scanning FOVT positions for consecutive captures of the Tele images;capturing a Tele image at each respective scanning FOVT position;aligning the captured Tele images with segments of the R image to obtain aligned Tele images; andusing the aligned Tele images and the R image to create a new image having a field of view FOVN≤FOVR, wherein the image resolution of the new image RESN>RESR and/or wherein the SNR of the new image SNRN>SNRR, wherein the determining an order of one or more scanning FOVT positions is performed so that the composed new image covers a maximal FOV according to a mechanical limitation of the scanning.

2. The method of claim 1, further comprising aligning each Tele image with the R image immediately after its capture and prior to the capture of an immediately following Tele image, analyzing each Tele image for faults, and if faults are detected in the Tele image, re-capturing the Tele image at a same FOVT position, or, if faults are not detected in the Tele image, proceeding to capture an immediately following Tele image at a respective FOVT position.

3. The method of claim 2, wherein the faults are selected from the group consisting of motion blur, electronic noise, rolling shutter, defocus blur and incorrect image alignment or obstructions.

4. The method of claim 2, wherein the faults are mechanical faults.

5. The method of claim 1, further comprising analyzing the aligned Tele images for faults, and if faults are detected in a particular Tele image, re-capturing the particular Tele image at a same FOVT position, or, if faults are not detected, using the aligned Tele images and the R image to create the new image.

6. The method of claim 1, wherein the aligned Tele images and the R image are fed into an algorithm to create a super wide image having a field of view FOVSW, wherein a FOV segment within FOVR included in at least one FOVT of the captured Tele images has a field-of-view union-FOVT, and wherein union-FOVT<FOVSW≤FOVR.

7. The method of claim 6, wherein the determining an order of one or more scanning FOVT positions includes capturing an object in a Tele image with a specific FOVT to improve RES or SNR of a similar object included in FOVN but not included in the specific FOVT.

8. The method of claim 1, wherein the determining an order of one or more scanning FOVT positions is performed so that the new image covers a region of interest selected by a user.

9. The method of claim 1, wherein the determining an order of one or more scanning FOVT positions is performed so that the new image covers a region of interest defined by an algorithm.

10. The method of claim 1, wherein the determining an order of one or more scanning FOVT positions is performed so that each Tele images includes a scene segments having a specific depth range or includes a scene segments that does not exceed a specific depth threshold.

11. The method of claim 1, wherein the determining an order of one or more scanning FOVT positions is performed so that first moving objects are captured, and after the moving objects are captured, stationary objects are captured.

12. The method of claim 1, wherein the determining an order of two or more FOVT positions is performed so that Tele images including specific scene characteristics within their respective FOVT s may be captured consecutively, and wherein the scene characteristics may be visual data such as texture or physical data such as brightness, depth or spectroscopic composition of a scene.

13. The method of claim 1, wherein the folded Tele camera captures two or more Tele images at two or more respective FOVT positions within FOVR, wherein the determining an order of two or more scanning FOVT positions is performed so that a moving object is removed from a scene included in FOVR.

14. A method, comprising: providing a folded Tele camera configured to scan and capture a plurality of Tele images, each captured image having a Tele image resolution (REST), a Tele image signal-to-noise-ratio (SNRT) and a Tele field of view (FOVT);obtaining and analyzing a reference (R) image with a R field of view FOVR>FOVT having a R image resolution RESR<REST, and/or a R image with a signal-to-noise-ratio SNRR<SNRT;determining an order of one or more scanning FOVT positions for consecutive captures of the Tele images;capturing a Tele image at each respective scanning FOVT position;aligning the captured Tele images with segments of the R image to obtain aligned Tele images; andusing the aligned Tele images and the R image to create a new image having a field of view FOVN≤FOVR, wherein the image resolution of the new image RESN>RESR and/or wherein the SNR of the new image SNRN>SNRR, and wherein the determining an order of two or more FOVT positions is performed so that capturing a minimal number of Tele images is required.