Convolutional Neural Networks Approximating Image Processing Operators
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image processing technologies face inefficiencies in executing complex imaging operators, often requiring resolution reduction which results in reduced image quality, limiting their ability to enhance images effectively.
Innovation Solution
The use of convolutional neural networks (CNNs) to approximate complex image processing operators by training on input-output image pairs, allowing for efficient computation of image processing functions without running the original operators, and incorporating context aggregation through dilated convolutions to maintain high-resolution image quality.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If complex image processing operators are executed directly, then image quality is maintained, but computational resources are consumed significantly
Solution Approach 1:
The patent applies preliminary action by training a convolutional neural network in advance to learn and store the mapping relationships of complex image processing operators. During actual operation, the pre-trained CNN directly processes images without executing the original complex operators, thus maintaining image quality while dramatically reducing computational resource consumption during runtime.
2Productivity
If resolution reduction is applied to improve efficiency, then computational resources are reduced, but image quality deteriorates
Solution Approach 1:
The patent uses copying by creating a trained CNN model that replicates the functionality of complex image processing operators. This copied model processes images at full resolution without requiring downsampling, thereby maintaining image quality while achieving computational efficiency through the learned patterns in the neural network weights.
3Productivity
If existing efficiency improvement methods are applied, then computational resources are reduced, but image quality is compromised
Solution Approach 1:
The patent replaces the mechanical system of direct operator execution with a neural network-based system. The trained CNN substitutes the original complex image processing operators, enabling efficient computation through learned feature representations while preserving image quality through the network's ability to capture complex patterns without resolution reduction.
Data Source
AI summary
Techniques are provided for approximating image processing functions using convolutional neural networks (CNNs). A methodology implementing the techniques according to an embodiment includes performing, by a CNN, a sequence of non-linear operations on an input image to generate an output image. The generated output image approximates the application of a targeted image processing operator to the input image. The CNN is trained on pairs of training input and output images, wherein the training output images are generated by application of the targeted image processing operator to the training input images. The CNN training process generates bias parameters and convolutional kernel parameters to be employed by the CNN for processing of intermediate image layers associated with processing stages between the input image and the output image, each of the processing stages associated with one of the sequence of non-linear operations. The parameters are associated with the targeted image processing operator.