TECHNIQUES FOR RESAMPLING IMAGES

Number: 20230377093
Published: 2023-11-23
Filed: 2023-05-19
Assignee: DISNEY ENTERPRISES, INC.
Inventors: Djelouah; Abdelaziz et al.
CPC: G06T3/02; G06T3/18; G06T3/4046; G06T5/60; G06T5/73; G06T5/80
Verdict: Set aside generic image resampling
Source: Google Patents · FreePatentsOnline

Abstract

Techniques are disclosed for resampling images. In some embodiments, a resampling model includes (1) one or more feature extraction layers that extract features from an input image and a degradation map; (2) one or more resampling layers that generate warped features from the extracted features and a warp grid; and (3) one or more prediction layers that generate, from the warped features, an output image or resampling kernels that can be applied to the input image to generate an output image. In some embodiments, the resampling model can be trained by applying degradation maps to output images in a training data set to generate corresponding input images, and training the resampling model using the input images and the corresponding output images.

Background

BACKGROUND Technical Field

Embodiments of the present disclosure relate generally to computer science and computer graphics and, more specifically, to techniques for resampling images. Description of the Related Art

Image resampling is the process of geometrically transforming digital images. Examples of image resampling include upscaling an image to a higher resolution image and/or applying a transformation operation on the image, such as affine transform (e.g., translation, scaling, reflection, rotation, shearing, or the like) or a lensing or delensing transform that increases or decreases a spherical warping of the image. Many image processing techniques involve image resampling.

One conventional approach for resampling involves determining, for each of the pixels of an input image, corresponding pixels of an output image. For example, a 64-pixel-square input image can be linearly interpolated to distribute the channel values of each single pixel over a 2×2 region in order to produce a 128-pixel-square upsampled output image. One drawback of such an approach for resampling images is that the input image is sequentially processed through multiple different operations, which can introduce or amplify visual artifacts in the output image. As a result, the output image can have relatively poor overall visual quality.

As the foregoing illustrates, what is needed in the art are more effective techniques for resampling images. SUMMARY

One embodim

Claims

1. A computer-implemented method for resampling images, the method comprising: extracting one or more features from an input image and a degradation map; determining, based on the one or more features and a warp grid that maps one or more coordinates of one or more pixels in an output image to one or more corresponding coordinates in the input image, one or more warped features; and generating at least one of the output image or one or more resampling kernels based on the one or more warped features. || 10. One or more non-transitory computer-readable storage media including instructions that, when executed by one or more processing units, cause the one or more processing units to perform steps for resampling images, the steps comprising: extracting one or more features from an input image and a degradation map; determining, based on the one or more features and a warp grid that maps one or more coordinates of one or more pixels in an output image to one or more corresponding coordinates in the input image, one or more warped features; and generating at least one of the output image or one or more resampling kernels based on the one or more warped features. || 18. A computer-implemented method of training a resampling model to generate resampled images, comprising: generating one or more warped images based on one or more images and a degradation map; and performing one or more operations to train the resampling model based on the one or more images and the one or more warped images.