Image Resolution Enhancement via Supervised Pattern Classification
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for scaling low-resolution images to high-resolution images, such as those described in U.S. Pat. No. 6,058,248, often result in blurry images due to the creation of unwanted visual artifacts like false edges and blurred features, caused by unsupervised training procedures that lead to dense or sparse cluster centers.
Innovation Solution
A method that selects low-resolution pixels, generates feature vectors to characterize patches, classifies these patches into top-ranking patterns, and applies specific filters or interpolation processes to generate high-resolution pixels, using a large dataset of high-resolution sample images to determine filters that reduce content dependency and improve image quality.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Extent of automation
If unsupervised learning is used to create filters from sample data, then the method can automatically determine filters without manual intervention, but the resulting images contain unwanted visual artifacts such as false edges and blurred features
Solution Approach 1:
The patent introduces a feedback mechanism by using ground truth high-resolution images to supervise the filter learning process. The supervised learning approach compares the generated high-resolution images with the actual ground truth images and adjusts the filters accordingly, preventing the generation of false edges and blurred features while maintaining automatic operation.
2Adaptability or versatility
If a large number of filter classes are created to cover all possible image patterns, then the method can handle diverse image content, but the computational complexity and storage requirements increase significantly
Solution Approach 1:
The patent changes the parameter of filter classification by using a fixed, small number of filter classes (e.g., 5 filters) instead of creating a large number of classes to cover all image patterns. This approach simplifies the filter database while maintaining effectiveness through carefully selected filters that can handle diverse image content.
3Productivity
If conventional interpolation methods are used to scale low-resolution images, then the process is simple and fast, but the output images appear blurry with lost detail
Solution Approach 1:
The patent applies preliminary action by pre-computing and storing optimal filters through supervised learning on a large dataset of high-resolution images. During actual image scaling, these pre-learned filters are applied directly without requiring complex real-time computation, achieving both speed and quality improvement over conventional interpolation methods.
Data Source
AI summary
A method for improving image resolution includes (a) selecting a low-resolution pixel in a low-resolution image; (b) generating a vector that characterizes a low-resolution patch about the low-resolution pixel; (c) classifying the low-resolution patch into one of a plurality of possible patterns; (d) if said one pattern is one of a plurality of top ranking patterns having filters, applying a filter of said one of the top ranking patterns to the low-resolution patch to generate high-resolution pixels that correspond to the low-resolution pixel; (e) if said one pattern is not one of the plurality of top ranking patterns, applying an interpolation process to the low-resolution patch to generate the high-resolution pixels; (f) repeating steps (a) to (e) for additional low-resolution pixels to generate additional high-resolution pixels to form a high-resolution image.