Patch-Based Image Classifier Training Reducing Computational Resources
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current object detection, recognition, and classification methods in computer vision are not scalable and are computationally intensive due to reliance on supervised data and localized image features like Histogram of Oriented Gradients (HOG), making them inefficient for real-world applications with diverse visual appearances.
Innovation Solution
The approach leverages patch-based features extracted from weakly supervised images to optimize multi-class image classification, reducing computational resources and improving efficiency and accuracy by training classifiers using patch-based features and global features.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If supervised data and localized image features (e.g., HOG) are used for object detection and classification, then classification accuracy can be achieved, but computational resources and time consumption increase significantly
Solution Approach 1:
The patent segments the image into multiple patches and processes them in parallel using multiple classifiers, rather than processing the entire image through a single complex classifier. This division allows computational work to be distributed and executed concurrently, reducing overall processing time while maintaining classification accuracy through ensemble voting of multiple classifiers.
Solution Approach 2:
The patent uses multiple classifiers (excessive action) to process different patches of the image simultaneously. By deploying more classifiers than a single traditional approach would use, the system achieves parallel processing that reduces total computational time, with the ensemble vote providing robust classification results.
2Measurement precision
If supervised data and localized image features are used for training classifiers, then object recognition accuracy improves, but scalability is reduced due to high computational intensity
Solution Approach 1:
The patent divides the training and processing workload into independent patch-level operations that can be scaled horizontally. Each patch can be processed by independent classifiers, allowing the system to scale by adding more classifiers and processors rather than increasing the complexity of a single classifier, thus improving scalability while maintaining recognition accuracy.
Solution Approach 2:
The patent creates a universal framework where multiple simple classifiers can perform the same function on different patches. This multi-functionality allows the system to scale by replicating the same classifier architecture across multiple processing units, improving productivity and scalability without sacrificing recognition accuracy through the ensemble approach.
3Adaptability or versatility
If current object detection methods are applied to handle diverse visual appearances in real-world applications, then comprehensive object recognition can be achieved, but computational resource requirements become prohibitive
Solution Approach 1:
The patent segments the complex task of recognizing diverse objects into simpler patch-level classification tasks. Each classifier processes only local features of patches, requiring less computational energy individually. The segmentation allows the system to handle diverse visual appearances by combining results from multiple energy-efficient local classifiers rather than using one energy-intensive global classifier.
Solution Approach 2:
The patent employs multiple partial classifiers (each handling only specific patches) rather than one complete classifier processing the entire image. This excessive use of simple partial classifiers reduces the computational energy each unit consumes, while the collective ensemble maintains the adaptability to handle diverse visual appearances across the whole image.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
Optimizing multi-class image classification by leveraging patch-based features extracted from weakly supervised images to train classifiers is described. A corpus of images associated with a set of labels may be received. One or more patches may be extracted from individual images in the corpus. Patch-based features may be extracted from the one or more patches and patch representations may be extracted from individual patches of the one or more patches. The patches may be arranged into clusters based at least in part on the patch-based features. At least some of the individual patches may be removed from individual clusters based at least in part on determined similarity values that are representative of similarity between the individual patches. The system may train classifiers based in part on patch-based features extracted from patches in the refined clusters. The classifiers may be used to accurately and efficiently classify new images.