Adaptive Object Identification via Dynamic Local Region Selection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing object identification techniques face challenges in accurately identifying individuals from facial images due to variations in pixel values caused by orientation, expression, and illumination conditions, making it difficult to select the optimal number of local regions for effective identification.
Innovation Solution
An image processing apparatus and method that calculates the degree of similarity between local feature amounts in an input image and registered images, considers the capturing conditions, and selects the appropriate local similarities to determine whether the input image belongs to the same category as the registered image, by adjusting the number of local regions based on the variation in shooting conditions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If a predetermined number of local regions are selected in descending order of similarity, then identification can be implemented with limited computational resources, but identification accuracy deteriorates when shooting conditions vary significantly
Solution Approach 1:
The patent applies dynamics by making the number of local regions to be collated adaptive rather than fixed. The collation process dynamically adjusts the number of local regions based on the shooting condition variation detected during comparison. When shooting conditions are similar, more local regions are collated to improve accuracy; when conditions differ significantly, fewer local regions are used to maintain processing efficiency. This dynamic adjustment resolves the contradiction between processing speed and identification accuracy.
2Measurement precision
If more local regions are collated, then identification accuracy improves under similar shooting conditions, but processing time increases
Solution Approach 1:
The patent applies parameter changes by varying the number of local regions to be collated based on the shooting condition parameter. Instead of using a fixed number of local regions, the system changes this parameter dynamically according to the detected variation in shooting conditions. This allows the system to optimize the balance between processing time and identification accuracy by adjusting the collation scope appropriately for each comparison task.
3Reliability
If local regions with high similarity are selected, then variations in pixel values due to orientation and expression are reduced, but the optimal number of regions varies with shooting condition variation
Solution Approach 1:
The patent applies dynamics by making the collation process adaptive to different shooting conditions. The system detects the variation in shooting conditions and dynamically adjusts the number of local regions to collate accordingly. This dynamic adaptation allows the system to maintain robustness against variations while being versatile enough to handle different shooting scenarios optimally.
Solution Approach 2:
The patent applies parameter changes by adjusting the number of local regions to be collated based on the shooting condition parameter. When shooting conditions are similar, the system increases the number of local regions for collation to improve reliability. When conditions vary significantly, it reduces the number to maintain adaptability and processing efficiency. This parameter adjustment resolves the contradiction between robustness and adaptability.
Data Source
AI summary
The degree of similarity between corresponding local feature amounts out of a plurality of local feature amounts of the object in the input image and a plurality of local feature amounts of an object in an image registered in advance is obtained. At least one degree of similarity is selected out of the obtained degrees of similarity based on a capturing condition for the object in the input image and a capturing condition for the object in the registered image, and one degree of similarity is derived from the at least one selected degree of similarity. It is determined based on the one derived degree of similarity whether the object in the input image belongs to the same category as the object in the registered image.