Convolutional Feature Fusion for Image Processing Performance
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Convolution neural networks (CNNs) are limited in reflecting new characteristics not obtainable from previous feature information during image processing, as they rely solely on feature information generated in previous layers for convolution operations.
Innovation Solution
An image processing apparatus and method that combines feature information from characteristic information generated through convolution operations with existing feature information, using this combined information for subsequent convolution operations to enhance image processing performance.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If convolution operation uses only feature information from previous layer, then device complexity is reduced, but image processing performance deteriorates due to inability to reflect new characteristics
Solution Approach 1:
The patent segments the feature information processing by separating it into two distinct paths: (1) conventional feature information from previous convolution layers, and (2) new characteristic information obtained through additional filtering operations. This segmentation allows the system to process different types of information through appropriate mechanisms, improving overall processing performance while managing complexity through structured organization of the neural network architecture.
Solution Approach 2:
The patent introduces a new dimension of information processing by adding filtering operations that generate characteristic information beyond the conventional feature maps. This dimensional expansion allows the neural network to access and utilize additional characteristics (such as frequency domain information or other transformed features) that are not obtainable from standard convolutional features alone, thereby improving image processing performance without requiring a complete redesign of the base architecture.
2Reliability
If new characteristic information is obtained through additional filtering operations, then image processing performance is improved, but device complexity increases
Solution Approach 1:
The patent applies preliminary action by performing filtering operations on feature information to extract characteristic information before it is fed into subsequent convolutional layers. This preprocessing step prepares the data in advance, enabling the main convolutional processing to utilize enriched feature representations that include both original and transformed characteristics, thereby improving performance while keeping the main processing path relatively simple.
Solution Approach 2:
The patent introduces filtering operations as an intermediary mechanism that transforms raw feature information into enhanced characteristic information. This intermediary processing layer acts as a bridge between conventional feature extraction and subsequent convolutional operations, allowing the system to incorporate additional characteristics (such as frequency information or other domain-transformed features) without directly modifying the core convolutional architecture, thus managing complexity effectively.
Data Source
AI summary
Provided is an image processing apparatus including a memory storing at least one instruction, and a processor configured to execute the at least one instruction stored in the memory to obtain first feature information by performing a convolution operation on a first image and a first kernel included in a first convolution layer among a plurality of convolution layers, obtain at least one piece of characteristic information, based on the first feature information; obtain second feature information, based on the first feature information and the at least one piece of characteristic information, obtain third feature information by performing a convolution operation on the obtained second feature information and a second kernel included in a second convolution layer that is a layer next to the first convolution layer among the plurality of convolution layers, and obtain an output image, based on the third feature information.