Image Processing Apparatus Texture Restoration via Weighted Random Patches
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current image processing technologies fail to effectively restore the texture component of images lost due to image enlargement and compression, leading to degraded image fineness, especially in transitioning from low definition to high definition formats like 4K or 8K UHD.
Innovation Solution
An image processing apparatus and method that utilizes random patches with pseudo random numbers, where correlations between pixel blocks and random patches are calculated to obtain weights, which are applied to generate an output image by adding weighted random patches to pixel blocks, thereby restoring texture and improving image fineness.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of energy
If image compression techniques are applied to reduce data size, then storage and transmission efficiency is improved, but texture component is lost and image fineness is degraded
Solution Approach 1:
The patent extracts the texture component from the compressed image by analyzing frequency domain information and identifying regions where texture has been lost during compression. This allows selective restoration of only the texture portions without reprocessing the entire image, thus maintaining efficiency while improving fineness.
Solution Approach 2:
The patent changes parameters by transforming the image to frequency domain, analyzing spectral characteristics, and applying targeted filtering operations. By modifying frequency domain parameters and applying inverse transformation, the texture component is restored without significantly increasing data size or processing overhead.
2Manufacturing precision
If conventional image processing methods are used to restore texture, then image fineness is improved, but side effects like ringing artifacts occur
Solution Approach 1:
The patent applies local quality by performing texture restoration only in specific regions where texture loss is detected, rather than uniformly processing the entire image. By identifying texture-lost regions through frequency analysis and applying restoration operations selectively, the method improves fineness while avoiding ringing artifacts in non-affected areas.
Solution Approach 2:
The patent uses partial action by applying texture restoration only to the extent necessary - specifically to regions where texture has been lost. The processing is neither too weak (which would fail to restore texture) nor too strong (which would cause ringing artifacts), achieving optimal balance through adaptive region-based processing.
3Manufacturing precision
If random patches with pseudo random numbers are generated and applied to pixel blocks, then texture component is restored and image fineness is improved, but memory usage increases
Solution Approach 1:
The patent uses copying by generating random patches that replicate texture patterns and applying them to pixel blocks where texture is lost. Instead of storing complete high-resolution texture data, the system generates and applies compact random patch representations, significantly reducing memory requirements while restoring texture fineness.
Solution Approach 2:
The patent applies preliminary action by pre-generating random patches with specific statistical properties (zero mean, controlled variance) before they are needed for texture restoration. These pre-prepared patches are then efficiently applied to pixel blocks during processing, reducing both computation time and memory usage during the actual restoration operation.
Data Source
AI summary
Disclosed is an image processing apparatus and a method of operating the same. The image processing apparatus includes: a memory storing information on at least one random patch; and at least one processor configured to: obtain correlations between a pixel block included in an input image and each of a plurality of random patches obtained from the information on the at least one random patch, obtain weights respectively for the plurality of random patches on a basis of the obtained correlations and apply the weights respectively to the plurality of random patches, and obtain an output image by applying, to the pixel block, the plurality of random patches to which the weights are respectively applied.