Icon Identification via Semantic Feature Extraction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing object detection systems fail to accurately identify new or dissimilar objects within images due to reliance on fixed training sets, leading to false positives and unsatisfactory results, especially in real-world scenarios with varying backdrops, illumination, and geometric transformations.
Innovation Solution
A hybrid pipeline approach using deep learning techniques for object identification, where a deep learning model extracts semantic features from images and compares them to known objects, allowing for accurate detection and classification of new objects without additional training, and incorporates a finer-grained search approach to handle variations.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If a fixed training set of manually labeled images is used for object detection, then the system can achieve reliable detection for objects within the training set, but it fails to accurately identify new or dissimilar objects and produces false positives
Solution Approach 1:
The patent transforms the static object detection problem into a dynamic one by allowing the system to adapt its detection criteria based on query images. Instead of relying on a fixed training set, the system dynamically adjusts its matching thresholds and parameters based on the specific query being evaluated, enabling it to handle new and dissimilar objects effectively
Solution Approach 2:
The patent introduces an intermediary verification process that acts as a bridge between the query image and the object database. This intermediary system performs pairwise comparisons and uses verification images to mediate the detection process, allowing the system to accurately identify new objects without requiring them to be in the original training set
2Stability of the object's composition
If traditional object detection systems are trained on fixed datasets, then they achieve consistent results for known objects, but they require extensive retraining and cannot handle real-world variations in illumination, backdrop, and geometric transformations
Solution Approach 1:
The patent performs preliminary actions by pre-processing query images and verification images to normalize them before comparison. This includes adjusting for illumination variations, geometric transformations, and backdrop differences in advance, so that the actual object matching process can focus on intrinsic features rather than being distracted by environmental variations
Solution Approach 2:
The patent dynamically changes detection parameters such as matching thresholds, confidence levels, and comparison criteria based on the specific characteristics of the query image and the expected object variations. This allows the system to maintain stable detection for known objects while adapting to handle variations in illumination, backdrop, and geometry
Data Source
AI summary
Systems, devices, media, and methods are presented for graphical icon identification within an image or video stream. The systems and methods receive an image including a graphical icon. The systems and methods identify a set of proposed regions of the image, at least one proposed region of the set of proposed regions containing the graphical icon and extract a set of semantic features for each proposed region of the set of proposed regions. Based on the set of semantic features of the set of proposed regions, the systems and methods identify a set of proposed icons corresponding to the graphical icon included in the image and determine a match between the graphical icon and at least one proposed icon of the set of proposed icons.