The present disclosure is related to systems and methods for noise reduction. The method includes obtaining a current frame including a plurality of first pixels. The method includes determining an interframe difference between each first pixel in the current frame and a corresponding pixel in a previous frame obtained prior to the current frame. The method includes generating a denoised frame by performing a first noise reduction operation on the current frame. The method includes determining an intraframe difference for each second pixel in the denoised frame. The method includes generating a target frame by performing a second noise reduction operation on the denoised frame.

Apparatuses, systems, and techniques are presented to reconstruct one or more images. In at least one embodiment, one or more neural networks are used to determine one or more blending weights for one or more images based, at least in part, upon one or more pixel value masks for the one or more images.

This application provides an image processing method, apparatus, a device, and the like. In the method, a special image filter is generated, and super-resolution is performed on an image based on the image filter, thereby improving an image super-resolution effect. The image filter includes filter parameters corresponding to each pixel in an image that requires super-resolution processing, and pixels with different texture features correspond to different filter parameters. The image super-resolution method, apparatus, device, and the like may be applied to various scenarios such as a video, a game, and photographing, to improve an image effect in these scenarios, and enhance user experience.

Apparatuses, systems, and techniques are presented to train one or more neural networks. In at least one embodiment, one or more neural networks are trained based, at least in part, on one or more image sequences, where backpropagation is performed using one or more subsets of images from the one or more image sequences.

A system for super-sampling digital images detects artifacts in an SRGAN super-sampled image, determines blocks of the image that contribute to the artifacts, and if the artifact-contributing blocks exceed a threshold, discards the SRGAN generated output image in favor of applying a super-sampled image generated by an alternate mechanism, such as a nearest neighbor algorithm.

A camera system captures an image in a source aspect ratio and applies a transformation to the input image to scale and warp the input image to generate an output image having a target aspect ratio different than the source aspect ratio. The output image has the same field of view as the input image, maintains image resolution, and limits distortion to levels that do not substantially affect the viewing experience. In one embodiment, the output image is non-linearly warped relative to the input image such that a distortion in the output image relative to the input image is greater in a corner region of the output image than a center region of the output image.

Methods, devices and non-transitory computer-readable storage medium for processing a sequence of respective top view images of a same terrestrial location are provided. One of the method may comprise choosing one image, called reference image, among the respective top view images, estimating for each respective top view image a respective geometric deformation between the respective top view image and the reference image, computing by the respective geometric deformations respective subpixel positions of the respective top view images relative to one high-resolution coordinate system, interpolating at the respective subpixel positions to sample at least part of at least some of the respective top view images on a prescribed grid to obtain a high-resolution image.

A method for processing an image includes: an image to be processed with a first resolution is acquired; and the image to be processed is processed by a target neural network model to obtain a target image, the target image being a denoised image with a second resolution, the second resolution being higher than the first resolution, and the target neural network model including a first preset number of convolutional layers and a second preset number of sub-pixel up-sampling portions.

An image display method, a display system, and a computer-readable storage medium are disclosed. The image display method includes: acquiring a first image; determining a first region and a second region in the first image; performing a first rendering algorithm on the first region in the first image and performing a second rendering algorithm on the second region in the first image, so as to obtain a second image, where a rendering resolution of the first rendering algorithm is greater than a rendering resolution of the second rendering algorithm; and displaying the second image.

A computer-implemented method for training a generator to manipulate one or more characteristics of an image is disclosed. The method comprises training a generator to output warp fields that modify one or more characteristics of an image, wherein training the generator comprises use of a Generative Adversarial Network (GAN) and training data comprising a plurality of images.