Multi-patch Super-resolution Using Scale-invariant Self-similarity
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing super-resolution methods are computationally intensive due to the need for searching high-resolution counterparts in databases or dictionaries, making them challenging for commercial applications, especially when relevant examples are scarce and hardware implementation is inefficient.
Innovation Solution
The proposed method employs a scale-invariant self-similarity (SiSS) based approach that selects patches within the image itself, using multi-shaped and multi-sized patches to reconstruct high-resolution images without the need for database searches, and incorporates a hybrid weighting method to suppress artifacts, significantly reducing computational complexity.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If example-based SR methods use database searching to reconstruct HR images, then image quality can be improved, but computational complexity increases significantly
Solution Approach 1:
The patent extracts only the necessary self-similarity information directly from the input LR image itself, rather than searching external databases. By taking out and utilizing the inherent self-similarity characteristics within the image, the method eliminates the need for complex database searches while maintaining the ability to reconstruct HR details.
Solution Approach 2:
The method enables the LR image to serve itself by exploiting its own self-similarity characteristics. The image contains redundant information within itself that can be used for reconstruction, eliminating the need for external databases or complex searching mechanisms.
2Measurement precision
If larger databases are used in example-based SR methods, then image reconstruction quality improves, but time and memory consumption increase
Solution Approach 1:
The method uses the input image itself as the source of reconstruction information, eliminating the need for external databases. The self-similarity characteristics within the image provide all necessary information for HR reconstruction, avoiding the time and memory costs of storing and searching large databases.
Solution Approach 2:
The single input LR image serves multiple functions: it is both the source image to be enhanced and the reference database for finding self-similar patches. This multi-functionality eliminates the need for separate training databases while providing sufficient information for reconstruction.
3Reliability
If traditional SR methods perform patch searching in databases, then relevant examples can be found for reconstruction, but the process becomes computationally intensive
Solution Approach 1:
The image serves itself by providing its own self-similar patches for reconstruction. This eliminates the need for searching external databases, dramatically reducing computational complexity while maintaining reconstruction accuracy through the inherent self-similarity of natural images.
Solution Approach 2:
The method performs preliminary organization of patch information by exploiting the inherent self-similarity structure within the image before reconstruction. This preliminary organization eliminates the need for computationally intensive searching during the reconstruction phase.
Data Source
AI summary
Embodiments of the present invention include apparatuses, systems and methods for multi-patch based super-resolution from a single video frame. Such embodiments include a scale-invariant self-similarity (SiSS) based super-resolution method. Instead of searching HR examples in a database or in LR image, the present embodiments may select the patches according to the SiSS characteristics of the patch itself, so that the computational complexity of the method may be reduced because there is not any search involved. To solve the problem of lack of relevant examples in natural images, the present embodiments may employ multi-shaped and multi-sized patches in HR image reconstruction. Additionally, embodiments may include steps for a hybrid weighing method for suppressing artifacts. Advantageously, certain embodiments of the method may be 10˜1,000 times faster than the example based SR approaches using patch searching and can achieve comparable HR image quality.