Deconvolution Kernel Segmentation for Image Processing Efficiency
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional image processing de-convolution methods require significant storage space and reduce efficiency due to the need for tic-tac-toe filling with zeros, which involves invalid weight values in the convolution operation.
Innovation Solution
The method splits the de-convolution kernel into sub-convolution kernels, allowing only valid weight values to participate in the convolution operation, eliminating the need for zero filling and reducing storage and calculation requirements.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Volume of moving object
If tic-tac-toe filling with zeros is performed on the input image feature matrix, then the image size is enlarged, but storage space is significantly occupied and calculation amount increases
Solution Approach 1:
The de-convolution kernel is segmented into multiple sub-convolution kernels based on the stride parameters. Each sub-convolution kernel corresponds to a specific position and processes only the necessary portion of the input feature matrix, eliminating the need for tic-tac-toe filling with zeros while achieving the same upscaling effect.
Solution Approach 2:
The patent extracts and processes only the valid weight values from the de-convolution kernel, excluding the zero-padded regions. By identifying and processing only the necessary computational elements, the method avoids storing and computing with unnecessary zero values, thus reducing storage space and calculation amount.
2Volume of moving object
If tic-tac-toe filling with zeros is performed on the input image feature matrix, then the image size is enlarged, but operation efficiency is significantly reduced
Solution Approach 1:
The de-convolution kernel is divided into multiple sub-convolution kernels, each handling a specific spatial position. This segmentation allows parallel processing of different regions of the feature matrix, avoiding the sequential zero-filling and convolution process, thereby improving operation efficiency.
Solution Approach 2:
Instead of performing convolution operations on the entire zero-padded matrix, the patent applies partial action by processing only the valid regions corresponding to non-zero kernel elements. This reduces the total number of multiplication and addition operations, significantly improving computational efficiency.
3Ease of manufacture
If de-convolution kernels are used with tic-tac-toe filling, then convolution operation can be performed, but invalid weight values participate in calculation increasing calculation amount
Solution Approach 1:
The patent extracts only the valid weight values from the de-convolution kernel by identifying the non-zero elements and their corresponding positions. These extracted valid weights are used to create sub-convolution kernels that process only the necessary input features, eliminating waste of computational energy on invalid zero-value operations.
Solution Approach 2:
The patent changes the parameters of the convolution operation by using stride-based indexing to directly access only the necessary elements in the input feature matrix. This parameter change eliminates the need for zero-filling and ensures that only valid weight values participate in the calculation, reducing calculation amount while maintaining the ease of convolution operation.
Data Source
AI summary
An image processing method is configured to split a deconvolution kernel according to a preset splitting mode to obtain a sub-convolution kernel. And then, determining an original sub-matrix corresponding to the sub-convolution kernel, according to parameters of the sub-convolution kernel and an image feature matrix, and performing a convolution operation on the original sub-matrix corresponding to the sub-convolution kernel by using the sub-convolution kernel to obtain a deconvolution sub-matrix corresponding to each sub-convolution kernel; determining a target feature matrix according to the deconvolution sub-matrix corresponding to the sub-convolution kernel. When performing the deconvolution operation by the method, the image feature matrix doesn't need to perform tic-tac-toe filling to reduce a storage space occupied in the deconvolution operation process; in addition, since zero filled in the tic-tac-toe isn't participated in the deconvolution operation, calculation amount can be greatly reduced, and calculation efficiency of the deconvolution operation can be improved.