Convolutional Neural Network Pixel Weighting for Image Processing
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing convolutional neural networks (CNNs) do not effectively reflect the reliability of neighboring pixels based on their distance from the center pixel during the convolution operation, leading to suboptimal image processing performance.
Innovation Solution
The proposed solution involves an image processing apparatus and method that uses a CNN to perform a convolution operation where the reliability of neighboring pixels is taken into account by determining first parameters based on the position information of pixels and samples, and then using these parameters to adjust weights or obtain sample values, thereby reflecting the reliability in the convolution process.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If the same weight is applied to all neighboring pixels in a convolution operation, then the calculation is simple and fast, but the reliability of neighboring pixels according to their distance from the center pixel is not reflected
Solution Approach 1:
The patent applies local quality by assigning different weights to neighboring pixels based on their distance from the center pixel. Specifically, pixels closer to the center pixel are given higher weights, while pixels farther away receive lower weights. This creates a weight map where each position has a locally optimized weight value that reflects its reliability, thereby resolving the contradiction between maintaining simplicity and improving reliability.
2Reliability
If distance-based weighting is implemented in convolution operations, then the reliability of neighboring pixels is reflected, but the computational complexity increases
Solution Approach 1:
The patent implements preliminary action by pre-calculating the weight map based on distance from the center pixel before performing the convolution operation. This weight map is computed once and then applied during convolution, avoiding the need to calculate distances and weights for each pixel during the actual convolution process. This approach reflects reliability information while reducing the computational burden during real-time processing.
3Productivity
If traditional convolution with uniform weights is used, then the processing speed is high, but the image processing performance is suboptimal
Solution Approach 1:
The patent applies parameter changes by modifying the weight parameter in the convolution operation from a uniform value to a distance-dependent value. Instead of using a single weight for all neighboring pixels, the system uses a weight parameter that varies based on the distance from the center pixel. This change in the weight parameter improves image processing performance by incorporating reliability information while maintaining the convolution framework.
Data Source
AI summary
An image processing apparatus may include: a memory storing one or more instructions; and a processor configured to execute the one or more instructions stored in the memory, wherein the processor executes the one or more instructions to use one or more convolutional neural networks to: determine first parameters, based on position information of first pixels included in a first region of a first image and position information of samples of the first image; obtain sample values of the samples by applying the first parameters to pixel values of the first pixels; obtain feature information corresponding to the first region by performing a convolution operation between the sample values and a kernel of the one or more convolutional neural networks; and generate a second image, based on the feature information.