Metadata-Guided Two-Stage Image Classification for Mobile Visual Search
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Mobile visual search applications on user devices are computationally intensive and time-consuming due to the need for analyzing images to identify features such as points of interest, which can be optimized by using metadata for initial classification and subsequent image analysis.
Innovation Solution
A method and apparatus that classify images by receiving metadata from a user device, performing an initial classification using only metadata to determine if the image is of a place, and a second classification using both metadata and image information, with optional parallel processing for place match filtering, and employing filter modules to select correct identifications.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If mobile visual search applications perform full image analysis to identify points of interest, then identification accuracy is improved, but computational intensity and time consumption increase
Solution Approach 1:
The image classification process is divided into two distinct stages: a first classification using only metadata (low computational cost) and a second classification using both metadata and image information (higher accuracy). This segmentation allows the system to quickly filter obvious cases while applying more intensive processing only when necessary, thereby reducing overall time consumption while maintaining identification accuracy.
Solution Approach 2:
The patent performs a preliminary classification using metadata before conducting the more computationally intensive image analysis. This preliminary action filters out cases that can be resolved with metadata alone, preventing unnecessary full image analysis and thus reducing time consumption while preserving accuracy for cases that require it.
2Measurement precision
If mobile visual search applications perform full image analysis to identify points of interest, then identification accuracy is improved, but computational intensity increases
Solution Approach 1:
The classification process is segmented into two stages with different computational requirements. The first stage uses only metadata requiring minimal computation, while the second stage uses both metadata and image information for higher accuracy. This segmentation reduces overall computational intensity by avoiding full image analysis for all queries while maintaining identification accuracy when needed.
Solution Approach 2:
The system performs partial image analysis (second classification) only when necessary, rather than applying full analysis to all images. This partial action approach reduces computational intensity by applying intensive processing selectively only to cases where metadata-based classification is insufficient, while maintaining identification accuracy for those specific cases.
3Reliability
If the system performs both first and second classification to ensure accurate place identification, then identification reliability is improved, but device complexity increases
Solution Approach 1:
The classification system is segmented into two sequential stages: first classification using metadata and second classification using both metadata and image information. This segmentation improves identification reliability by applying multiple levels of analysis while managing complexity through clear separation of processing stages, allowing each stage to have specialized, optimized processing logic.
Solution Approach 2:
The first classification using metadata serves as a preliminary action that filters and pre-processes images before the second classification. This preliminary action improves reliability by ensuring that only cases needing further analysis proceed to the second stage, while simplifying the overall system structure through a clear hierarchical processing flow.
Data Source
AI summary
Aspects of the disclosure pertain to identifying whether or not an image from a user's device is of a place or not. As part of the identification, a training procedure may be performed on a set of training images. The training procedure includes performing measurements of image data for each image in the set to derive a result. The result includes a series of variables for each training image in the set. The series of variable is evaluated for each training image to obtain one or more measurement weights and one or more measurement thresholds. These weights and thresholds are adjusted to set a false positive threshold and a false negative threshold for identifying whether an actual image is of a place type or is some other type of image.