Image Processing Device Using Patch-Based Dictionary Convolution
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The complexity of convolutional neural networks (CNNs) in image processing requires significant computation and memory, making them inefficient for practical applications.
Innovation Solution
An image processing device uses a dictionary generated by a deep neural network with n convolution layers, extracting feature information from input patches and performing convolution operations with filter kernels to reduce computational and memory requirements while maintaining performance.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If a deep neural network with multiple convolution layers is used for image processing, then image processing performance is improved, but computation and memory requirements increase significantly
Solution Approach 1:
The patent segments the deep neural network into multiple convolution layers and processes image patches through each layer sequentially. By dividing the image into patches and processing them through staged convolution operations, the system achieves complex image processing tasks while managing computational complexity through structured segmentation of both the input data and the processing architecture
Solution Approach 2:
The patent performs preliminary actions by pre-processing image patches through multiple convolution layers before final reconstruction. The dictionary learning and patch-based preprocessing steps are executed in advance to extract meaningful features, reducing the computational burden during the main processing phase and enabling efficient memory management
2Measurement precision
If traditional deep neural networks are used for image processing, then processing accuracy is maintained, but the amount of computation and memory required is excessive
Solution Approach 1:
The patent extracts meaningful features from image patches by performing convolution operations only on relevant local regions rather than processing entire images through the full neural network. By extracting and processing patches independently through convolution layers, the system maintains processing accuracy for critical features while significantly reducing the total computation and memory required compared to processing complete images
Solution Approach 2:
The patent applies partial action by processing only necessary image patches through the full convolutional pipeline, rather than processing every pixel or region with equal computational depth. The patch-based approach selectively applies complex convolution operations only where needed, maintaining accuracy for important features while reducing overall computational expenditure
Data Source
Figure 1
Figure 2
Figure 3
AI summary
The present disclosure relates to an image processing device including: a memory storing one or more instructions; a processor configured to execute the one or more instructions stored in the memory to: extract one or more input patches based on an input image; extract pieces of feature information respectively corresponding to the one or more input patches based on a dictionary including mapping information indicating mappings between a plurality of patches and pieces of feature information respectively corresponding to the plurality of patches; and obtain a final image by performing a convolution operation between the extracted pieces of feature information and a filter kernel, wherein the final image is an image obtained by performing image processing on the input image according to a preset purpose.