Super-Resolution Image Processing via Hash-Based Filter Bank Selection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current super-resolution image processing methods, particularly those involving up-sampling amplification, suffer from low processing speed due to the need to filter large numbers of pixels and frequent memory access, which affects the efficiency and response time during video conferences, live video streaming, and network course teaching.
Innovation Solution
The method involves acquiring hash values of low-resolution image pixels, querying pre-stored filter banks, filtering pixel regions using corresponding filters, and writing filtered values into high-resolution image coordinates, thereby reducing the number of filtered pixels and minimizing memory access by reading filter banks in bulk.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If filtering is performed on the up-sampled image with many pixels, then the super-resolution processing can be completed, but the processing speed becomes relatively low
Solution Approach 1:
The patent divides the filtering operation by first segmenting the low-resolution image into pixel regions, calculating hash values for each region, and then selectively filtering only the regions that need processing. This segmentation approach reduces the total number of pixels that require filtering, thereby improving processing speed while maintaining super-resolution quality.
Solution Approach 2:
The patent performs preliminary actions by pre-calculating hash values for pixel regions and pre-storing filter banks in memory. This allows the system to quickly identify which regions need filtering and retrieve the appropriate filters without performing unnecessary filtering operations on all pixels, thus enhancing processing efficiency.
2Measurement precision
If frequent memory access is performed to read filters for each pixel, then the filtering can be performed accurately, but the response speed is reduced
Solution Approach 1:
The patent pre-stores filter banks in memory during the initialization phase. This preliminary action allows the system to quickly retrieve filters when needed without performing frequent memory accesses during the actual filtering process, thereby reducing response time while maintaining filter accuracy.
Solution Approach 2:
The patent merges multiple filter retrieval operations into a single memory access by pre-loading filter banks into memory. This combining approach reduces the frequency of memory accesses and improves response speed while ensuring that the correct filters are retrieved for accurate processing.
Data Source
AI summary
Embodiments of the present disclosure provide an image processing method, device and apparatus, an electronic device, a computer-readable storage medium, a computer program product, and a computer program. The method comprises: obtaining a hash value of a pixel of a low-resolution image; finding a filter bank corresponding to the hash value; respectively filtering the regions corresponding to the pixels in the low-resolution image according to multiple filters in the filter bank such that each filter outputs a filtered pixel value; and writing each filtered pixel value into a corresponding coordinate position of a high-resolution image, thereby realizing super-resolution processing of the image and obtaining a high-resolution image. In the present disclosure, since the pixels in the low-resolution image are filtered, the number of pixels filtered is lower, and the processing efficiency is significantly improved.