Image processing using a convolutional neural network

Number: 10706503
Published: 2020-07-07
Filed: 2018-03-13
Assignee: Disney Enterprises, Inc.
Inventors: Schroers; Christopher, Perazzi; Federico, Hazirbas; Caner
CPC: G06T3/4046; G06T3/4053; G06T5/60; G06T5/70; G06T5/73
Verdict: Set aside generic CNN image processing, no creative hook
Source: Google Patents · FreePatentsOnline

Abstract

According to one implementation, an image processing system includes a computing platform having a hardware processor and a system memory storing a software code including a convolutional neural network (CNN) trained using one or more semantic map(s). The hardware processor executes the software code to receive an original image including multiple object images each identified with one of multiple object classes, and to generate replications of the original image, each replication corresponding respectively to one of the object classes. The hardware processor further executes the software code to, for each replication, selectively modify one or more object image(s) identified with the object class corresponding to the replication, using the CNN, to produce partially modified images each corresponding respectively to an object class, and to merge the partially modified images, using the CNN, to generate a modified image corresponding to the original image.

Background

BACKGROUND(1) Image processing techniques are important to many applications. For example, image restoration, in which an unknown image is recovered from a degraded version of itself, may utilize various image processing techniques to restore the unknown image. Despite its usefulness for recovering lost image features, image restoration presents significant challenges. Specifically, for instance, the problems posed by image restoration tend to be inherently underdetermined because multiple plausible images can be recovered from the same degraded original image.(2) Due to its underdetermination, image restoration requires some prior information about the image undergoing restoration. Traditional approaches to obtaining such prior information have been variously based on edge statistics, sparse representation, gradients, self-similarities, or some combination of those features. Nevertheless, there remains a need in the art for an image processing solution capable of using non-traditional forms of prior information to more effectively guide the accurate restoration or enhancement of a degraded or corrupted image.SUMMARY(3) There are provided systems and methods for performing image processing using a convolutional neural network, substantially as shown in and/or described in connection with at least one of the figures, and as set forth more completely in the claims.

Claims

1. An image processing system comprising: a computing platform including a hardware processor and a system memory; a software code stored in the system memory, the software code including a convolutional neural network (CNN) trained using at least one semantic map; the hardware processor configured to execute the software code to: receive an original image including a plurality of object images each identified with one of a plurality of object classes; generate a plurality of replications of the original image, each replication corresponding respectively to one of the plurality of object classes; for each replication, selectively modify at least one object image identified with the respective one corresponding object class, using the CNN, to produce a plurality of partially modified images corresponding respectively to the plurality of object classes; and merge the plurality of partially modified images, using the CNN, to generate a modified image corresponding to the original image. 11. A method for use by an image processing system including a computing platform having a hardware processor and a system memory storing a software code including a convolutional neural network (CNN) trained using at least one semantic map, the method comprising: receiving, using the hardware processor, an original image including a plurality of object images each identified with one of a plurality of object classes; generating, using the hardware processor, a plurality of replications of the original image, each replication corresponding respectively to one of the plurality of object classes; for each replication, selectively enhancing, using the hardware processor and the CNN, at least one object image identified with the respective one corresponding object class, to produce a plurality of partially modified images corresponding respectively to the plurality of object classes; and merging, using the hardware processor and the CNN, the plurality of partially modified images to generate a modified image corresponding to the original image.