Digital Image Orientation via Feature Extraction and Diverse Classifiers
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for automatically determining the orientation of digital images are inefficient and resource-intensive, particularly when dealing with large numbers of images, and there is a need for improved techniques that can accurately and quickly orient digital images to facilitate easier viewing and storage.
Innovation Solution
A method and system that extracts low-level features such as color coherence vectors and edge direction coherence vectors, and processes them using diverse classifiers like back-propagation neural networks and mixture of experts networks to determine the orientation of digital images, utilizing the YIQ color space for enhanced performance.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If existing methods for automatically determining orientation are used, then orientation detection can be achieved, but the process is inefficient and resource-intensive when dealing with large numbers of images
Solution Approach 1:
The patent segments the image processing task into distinct stages: extracting low-level features (color coherence vectors, edge direction coherence vectors), processing features through diverse classifiers (back-propagation neural networks, mixture of experts networks), and determining orientation. This segmentation allows each component to be optimized independently, improving overall efficiency while maintaining accuracy.
Solution Approach 2:
The patent transforms the image into the YIQ color space, changing the parameter representation from standard RGB to a color space that enhances the visibility of orientation-related features. This parameter transformation improves the effectiveness of subsequent feature extraction and classification operations, enabling faster and more accurate orientation detection.
2Measurement precision
If manual retrieval and orientation of digital images is performed, then accurate orientation can be ensured, but the process is very time-consuming and impractical for large numbers of images
Solution Approach 1:
The system enables automatic self-determination of image orientation through multiple diverse classifiers that independently analyze image features and vote on the correct orientation. This self-service mechanism eliminates the need for manual intervention while maintaining high accuracy through the collective decision-making of multiple classifiers.
Solution Approach 2:
The patent implements a feedback mechanism where diverse classifiers process image features and their outputs are combined to determine the final orientation. The system uses the combined output of multiple classifiers to make the final orientation decision, providing a form of feedback that improves accuracy compared to single-classifier approaches.
3Reliability
If simple orientation detection methods are used, then processing speed may be improved, but accuracy and robustness are reduced
Solution Approach 1:
The patent merges multiple diverse classifiers (back-propagation neural networks and mixture of experts networks) into a unified orientation determination system. Each classifier processes the same low-level image features but uses different algorithms, and their combined outputs provide a more robust and reliable orientation decision than any single classifier could achieve alone.
Solution Approach 2:
The system creates a composite classification approach by combining multiple diverse classifiers with different algorithmic characteristics. This composite structure leverages the strengths of each individual classifier type, creating a more robust and reliable orientation determination system that can handle various image types and conditions effectively.
Data Source
AI summary
A method of automatically determining orientation of a digital image comprises extracting features of the digital image and processing the extracted features using diverse classifiers to determine orientation of the digital image based on the combined output of the diverse classifiers.