Samsung Electronics Co., Ltd.

Texas Instruments Corporation

Other

Hamid Rahim Sheikh

John W Glotzbach

Ruiwen Zhen

Ibrahim Pekkucuksen

Long N Le

Michael Oliver Polley

John Seokjun Lee

Youngjun Yoo

Zeeshan Nadir

Yibo Xu

Yingmao Li

Nguyen Thang Long Le

Zhen Tong

Chenchi Luo

Kaimo Lin

Gyeongmin Choe

Raja Bala

George Q Chen

Tyler Luu

David D Liu

Sourabh Ravindran

Osman Gokhan Sezer

Kushal Kardam Vyas

Anqi Yang

Baekdu Choi

Deboleena Roy

Devendra Kumar Jangid

Manish Goel

Youngmin Kim

William M Johnson

Shashank Dabral

Rajasekhar Reddy Allu

Jeffrey L Nye

David Harold Bartley

Shalini Gupta

Murali S Chinnakonda

Akila Subramaniam

Ajay Jayaraj

Toshio Nagata

Gunawath Dilshan Godaliyadda

Stephen Busch

Rahul Rithe

John E Glotzbach

Omar A Elgendy

Yuting Hu

Pavan Chennagiri

Numair Khan

Mingzhao Yu

Weidi Liu

Abhiram Gnanasambandam

Jun Ki Cho

Seung-Chul Jeon

Robert Jp Nychka

Ganesh Sundararajan

John William Glotzbach

Ibrahim E Pekkucuksen

Seokjun Lee

A method includes identifying, using at least one processing device of an electronic device, a spatially-variant point spread function associated with an under-display camera. The spatially-variant point spread function is based on an optical transmission model and a layout of a display associated with the under-display camera. The method also includes generating, using the at least one processing device, a ground truth image. The method further includes performing, using the at least one processing device, a convolution of the ground truth image based on the spatially-variant point spread function in order to generate a synthetic sensor image. The synthetic sensor image represents a simulated image captured by the under-display camera. In addition, the method includes providing, using the at least one processing device, the synthetic sensor image and the ground truth image as an image pair to train a machine learning model to perform under-display camera point spread function inversion.

Machine learning model training using synthetic data for under-display camera (UDC) image restoration

A method includes obtaining an image and a gain map associated with the image. The method also includes identifying image patches in the image and corresponding gain map patches in the gain map. Different image patches are centered around different anchor points in the image. The method further includes, for each image patch and its corresponding gain map patch, generating an intensity-gain curve for the associated anchor point. The intensity-gain curve specifies (i) gain values based on the corresponding gain map patch for intensity values up to a threshold intensity value and (ii) gain values based on one or more input parameters for intensity values above the threshold intensity value. In addition, the method includes combining the intensity-gain curves to generate a 3D lookup table, which identifies the gain values for the anchor points in the image at each of multiple intensity values.

Generation of three-dimensional (3D) lookup table for tone mapping or other image processing functions

A system and method are provided for noise reduction improvement for blending blurred frames in a multi-frame system. The method includes retrieving a number of frames of an image. The method also includes identifying one or more edges within the frames and comparing an edge strength of a reference frame and an edge strength of a non-reference frame. The method further includes determining a weight reduction factor based on a result of the comparison and applying the weight reduction factor to a blending of multiple frames of the number of frames of the image. In addition, the method includes displaying the blended frames.

System and method for noise reduction for blending blurred frames in a multi-frame system

A method includes generating alignment maps for a first image frame having a first exposure level and a second image frame having a second exposure level different than the first exposure level. The method also includes generating, for the second image frame and a third image frame having a third exposure level different than the first and second exposure levels, shadow maps, saturation maps, and multi-exposure (ME) motion maps based on the alignment maps. The method further includes determining, based on the shadow maps, saturation maps, and ME motion maps, whether to perform machine learning-based denoising and, if so, on which image frame(s) to perform the machine learning-based denoising. In addition, the method includes updating at least one saturation map and at least one ME motion map for at least one of the second and third image frames depending on the image frame(s) on which the denoising is to be performed.

System and method for scene-adaptive denoise scheduling and efficient deghosting

A method includes aligning landmark points between multiple distorted images to generate multiple aligned images, where the multiple distorted images exhibit perspective distortion in at least one face appearing in the multiple distorted images. The method also includes predicting a depth map using a disparity estimation neural network that receives the multiple aligned images as input. The method further includes generating a warp field using a selected one of the multiple aligned images. The method also includes performing a two-dimensional (2D) image projection on the selected aligned image using the depth map and the warp field to generate an undistorted image. In addition, the method includes filling in one or more missing pixels in the undistorted image using an inpainting neural network to generate a final undistorted image.

System and method for facial un-distortion in digital images using multiple imaging sensors

A method includes obtaining a reference image frame and a non-reference image frame and generating a mask based on the reference image frame and the non-reference image frame. The method also includes determining a scaling coefficient and an offset coefficient for each of multiple color channels of the non-reference image frame based on pixels in the reference image frame and pixels in the non-reference image frame that are identified by the mask. The method further includes applying the scaling and offset coefficients to the non-reference image frame in order to generate a calibration-corrected non-reference image frame.

Dynamic calibration correction in multi-frame, multi-exposure capture

A method includes extracting multiple shallow features from a low-resolution image using a shallow feature extractor that includes a quaternion convolutional network. The method also includes extracting multiple deep features from the multiple shallow features using a deep feature extractor that includes multiple quaternion residual distillation blocks (QRDBs), where each QRDB includes a quaternion self-attention module. The method further includes reconstructing the multiple deep features into a high-resolution image. Each QRDB may further include a quaternion gated deconvolutional feed forward network (QGDFN) configured to suppress one or more of the multiple deep features.

System and method for single image super- resolution for smart device camera

A method includes dividing an image into overlapping image patches each having a specified size. The method also includes analyzing content of each image patch using a mathematical transform technique to classify each image patch into at least one class. The method further includes filtering each image patch for noise suppression by suppressing one or more transform coefficients of the image patch. An amount of suppression for each of the one or more transform coefficients is selected according to the at least one class of the image patch. In addition, the method includes reconstructing the filtered image patches into an output image.

System and method for adaptive discrete cosine transform (DCT) noise filtering for digital images

A method includes obtaining a raw image in a first image domain. The method also includes determining a color distribution and an amount of variation in the raw image. The method further includes using an iterative process to generate a dead leaf image from a blank image in the first image domain. The iterative process includes adding multiple circles and multiple sticks to the blank image until the dead leaf image is filled. The iterative process also includes blurring portions of the dead leaf image during at least one iteration of the iterative process. Textures of the multiple circles are blended based on the color distribution and the amount of variation in the raw image.

System and method for synthetic data generation using dead leaves images

A method includes filtering an input image through a discrete cosine transform based noise filter (DCT-NF). The DCT-NF converts the input image from an original space into a perceptual space, applying gamma correction. The DCT-NF separates luminance channels of the input image from chroma channels of the input image. The DCT-NF divides the input image into overlapping patches and computes DCT transform of the patches. Each patch is a different partial portion of the input image. The DCT-NF suppresses patches that include an input DCT coefficient within a threshold range. The DCT-NF applies an inverse discrete cosine transform (IDCT) to the suppressed patches and remaining overlapping patches that include an input DCT coefficient outside the threshold range. The DCT-NF re-combines luminance and chroma channels of the IDCT-transformed patches. The DCT-NF generates a DCT noise-filtered output image by re-converting the IDCT-transformed patches to the original space by applying inverse gamma correction.

Learning based discrete cosine noise filter

A method includes obtaining an input image that contains blur. The method also includes providing the input image to a trained machine learning model, where the trained machine learning model includes (i) a shallow feature extractor configured to extract one or more feature maps from the input image and (ii) a deep feature extractor configured to extract deep features from the one or more feature maps. The method further includes using the trained machine learning model to generate a sharpened output image. The trained machine learning model is trained using ground truth training images and input training images, where the input training images include versions of the ground truth training images with blur created using demosaic and noise filtering operations.

Machine learning-based approaches for synthetic training data generation and image sharpening

A method includes obtaining a blended red-green-blue (RGB) image frame of a scene. The method also includes performing, using at least one processing device of an electronic device, an interband denoising operation to remove at least one of noise and one or more artifacts from the blended RGB image frame in order to produce a denoised RGB image frame. Performing the interband denoising operation includes performing filtering of red, green, and blue color channels of the blended RGB image frame to remove at least one of the noise and the one or more artifacts from the blended RGB image frame. The filtering of the red and blue color channels of the blended RGB image frame is based on image data of at least one of the green color channel and a white color channel of the blended RGB image frame.

Apparatus and method for interband denoising and sharpening of images

A method includes obtaining, using at least one processing device of an electronic device, multiple image frames having different exposure levels. The method also includes determining whether a moving saturated region is present in at least some of the image frames. The method further includes selecting one of the image frames as a reference frame, where the selected image frame depends on whether the moving saturated region is present in at least some of the image frames. The selected image frame may have a first exposure level when the moving saturated region is present in at least some of the image frames, and the selected image frame may have a second exposure level longer than the first exposure level when the moving saturated region is not present in the image frames.

Reference frame selection for multi-frame image processing pipelines and other applications

An apparatus includes a memory configured to store images and a processor. The processor is configured to receive an input image. The processor is further configured to partition a human mask in the input image using a segmentation algorithm. The processor is also configured to generate a skin map based on identifying skin in the input image using the human mask. In addition, the processor is configured to process an output image with brightening applied using the skin map.

Automatic face and human subject enhancement algorithm for digital images

A method includes obtaining multiple image frames. The method also includes selecting, using at least one processing device of an electronic device, an asymmetrical image pair from the multiple image frames. The asymmetrical image pair includes a first image frame and a second image frame, where the first image frame has a shorter exposure than the second image frame. The method further includes identifying, using the at least one processing device, one or more features based on the asymmetrical image pair. The method also includes determining, using the at least one processing device, whether the first image frame contains flicker based on the one or more features. In addition, the method includes enabling or disabling, using the at least one processing device, the first image frame as a reference candidate based on the determination whether the first image frame contains flicker.

System and method for flicker avoidance in multi-frame multi-exposure image captures

Growing community of inventors

Allen, TX, United States of America

Hamid Rahim Sheikh