Image Restoration via Vector Conversion and Filtering
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional image restoration methods are inefficient due to high memory requirements and long processing times, especially when dealing with highly blurred images or high-resolution images, and often result in image discontinuities due to block division, and may not converge using methods like conjugate gradient or steepest descent.
Innovation Solution
The method converts two-dimensional image data into one-dimensional vectors, replaces matrix operations with image filtering, and uses numerical optimization methods like steepest descent or conjugate gradient to calculate gradient vectors, allowing for high-speed processing without block division and matrix operations.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If matrix operations are used for image restoration, then restoration accuracy is improved, but memory requirements and processing time increase significantly
Solution Approach 1:
The patent replaces matrix operations with convolution operations. Instead of using large matrix multiplications that require significant memory and computational resources, the invention uses convolution with kernel matrices that can be applied efficiently through image filtering techniques. This substitution maintains the ability to perform image restoration while dramatically reducing memory requirements and processing time.
Solution Approach 2:
The patent changes the approach from using large matrices to using small kernel matrices in convolution operations. By transforming the problem into the spatial domain and using small, fixed-size kernels that slide across the image, the system achieves similar restoration effects with much smaller computational parameters, thereby reducing memory requirements.
2Quantity of substance
If block division is used for high-resolution images, then memory capacity is managed, but image discontinuities occur at block boundaries
Solution Approach 1:
The patent applies segmentation by dividing the image into multiple blocks for processing. However, to maintain image continuity at block boundaries, the invention uses overlap regions where pixels in overlapping areas are processed together across block boundaries. This allows the system to manage memory by processing smaller blocks while preventing discontinuities through the overlap mechanism that ensures smooth transitions between blocks.
3Loss of time
If conventional edge enhancement is used, then processing time is short, but it cannot handle significant blurring
Solution Approach 1:
The patent creates a universal image restoration system that can handle various degrees of blurring and different image types using the same convolution-based framework. The system uses cost functions that can be adjusted for different blur conditions and employs multiple convolution operations with different kernels to address various restoration needs, making it universally applicable from mild to severe blurring without requiring separate specialized algorithms.
Solution Approach 2:
The patent performs preliminary actions by pre-calculating and storing multiple convolution kernels and cost function parameters before processing. This allows the system to quickly retrieve and apply appropriate restoration parameters without performing complex calculations during the actual restoration process, thereby maintaining short processing times while effectively handling significant blurring.
4Device complexity
If single numerical optimization method is used, then processing is simple, but convergence cannot be guaranteed
Solution Approach 1:
The patent implements dynamics by making the optimization process adaptive. The system dynamically selects between different numerical optimization methods (such as steepest descent and conjugate gradient) based on the specific characteristics of the image and blur conditions. This dynamic adaptation allows the system to choose the most effective method for each case, ensuring convergence while maintaining relatively simple processing through automated method selection.
Data Source
AI summary
The present application provides an image processing method of obtaining a clear image from a blurred image which includes the steps of converting an array of image data from a two dimension into a one-dimensional vector, the image data storing pixel values of the blurred image; converting all of two-dimensional matrix operations acting on the one-dimensional vector into an image filtering for the image data of the blurred image, and calculating values of statistic parameters necessary for calculation of a cost function estimating the clear image and a gradient vector of the cost function; and obtaining, by using the gradient vector being calculated, image data of the clear image by a numerical optimization method which calculates the image data of the clear image by using only the gradient vector such that a value of the cost function becomes maximum or minimum.