CNN Gradient Detection for Artifact-Free Image Upscaling
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing super-resolution (SR) systems, particularly those using generator networks like SRGANs, struggle to effectively upscale artificial images with computer-generated image gradients, often producing SR images with artifacts such as edges or contours that do not exist in the original image.
Innovation Solution
A system that utilizes a convolutional neural network (CNN) with residual and upscaling layers, along with classification layers to detect artificial images with computer-generated image gradients, and directs such images to an upscaling module for processing rather than relying solely on the CNN for upscaling, thereby avoiding the generation of unwanted artifacts.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If a generator network (SRGAN) is used to upscale images, then the upscaled images appear more realistic and detailed, but artifacts such as edges or contours are generated that do not exist in the original image
Solution Approach 1:
The system performs classification of the input image as artificial or natural before the upscaling operation. This preliminary action allows the system to select different processing paths: natural images are upscaled using the generator network for realistic results, while artificial images are routed to traditional upscaling methods to avoid generating artifacts. The classification step precedes and guides the upscaling decision.
Solution Approach 2:
The upscaling system is divided into separate processing paths: one path using the generator network for natural images and another path using traditional upscaling methods for artificial images. This segmentation allows each path to be optimized for its specific image type, preventing the generator network from processing artificial images where it would create harmful artifacts.
2Object-generated harmful factors
If a classification system is added to detect artificial images before upscaling, then artifacts are avoided in artificial images, but the system complexity increases
Solution Approach 1:
The classification network serves multiple functions: it identifies artificial images to prevent artifact generation, and simultaneously guides the routing decision for the upscaling process. This multi-functionality justifies the added complexity by providing dual benefits of artifact prevention and optimal processing path selection.
Solution Approach 2:
The classification network acts as an intermediary component between the input image and the upscaling processes. It analyzes the input image and determines the appropriate processing path, mediating between the two upscaling approaches (generator network and traditional methods) based on the image type.
Data Source
AI summary
An example system includes a processor and a non-transitory computer-readable medium having stored therein instructions that are executable to cause the system to perform various functions. The functions include obtaining an image associated with a print job, and providing the image as input to a convolutional neural network. The convolutional neural network includes a residual network, upscaling layers, and classification layers configured to detect whether the image is an artificial image having a computer-generated image gradient. The functions also include determining, based on an output of the classification layers, that the image is an artificial image having a computer-generated image gradient. Further, the functions include, based on determining that the image is an artificial image having a computer-generated image gradient, providing the image to an upscaling module of a print pipeline for upscaling rather than using an output of the upscaling layers for the upscaling.