Object Recognition Using Saliency and Location Context
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current object recognition systems struggle to accurately identify objects in images, especially when objects are rotated, blurred, or partially obstructed, and fail to distinguish between objects in photographs or video sequences.
Innovation Solution
A computer-implemented method that identifies objects in an image by using a database of identified objects, executing a saliency algorithm to determine salient objects, and determining a generic location for the image, which helps in identifying unidentified objects by querying databases with associated generic locations.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If feature-model mapping is used for object recognition, then the system can identify objects in standard conditions, but it fails to distinguish objects when they are rotated, blurred, or partially obstructed
Solution Approach 1:
The patent segments the object recognition process into multiple stages: initial feature-model mapping to identify candidate objects, saliency detection to highlight prominent objects, and context-based verification using generic location databases. This multi-stage segmentation allows the system to handle challenging conditions by progressively refining object identification beyond simple feature matching.
Solution Approach 2:
The patent introduces generic location databases and saliency maps as intermediary elements between the image input and final object identification. These intermediaries provide contextual information about expected objects in specific locations and highlight salient regions, enabling the system to overcome limitations of direct feature-model mapping when objects are rotated, blurred, or obstructed.
2Measurement precision
If multiple databases and algorithms are used to improve object identification accuracy, then confidence levels increase, but system complexity increases
Solution Approach 1:
The patent performs preliminary actions by pre-establishing generic location databases that associate objects with their typical spatial contexts, and by computing saliency maps that pre-identify prominent regions before final object recognition. These preliminary actions reduce the complexity of the final identification step by pre-processing contextual and visual prominence information.
Solution Approach 2:
The patent creates a multi-functional system where the generic location database serves multiple purposes: providing contextual expectations for object presence, verifying candidate object identifications, and guiding attention to relevant image regions. The saliency algorithm similarly serves both to highlight important objects and to guide the verification process, reducing overall system complexity through functional consolidation.
Data Source
AI summary
Identifying objects in an image. An image is received. One or more objects in the image are identified, based on a database of identified objects, and wherein one or more other objects in the image are unidentified based on the database of identified objects. One or more salient objects in the image is identified, based on execution of a saliency algorithm. A generic location for the image is determined, based on the one or more identified salient objects and a database that associates objects with generic locations. One or more of the unidentified objects are identified, based on the determined generic location for the image.