Facial Image Recognition via Occlusion Pattern Exclusion
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image recognition systems face reduced recognition accuracy when facial images are partially occluded by articles like glasses or masks, as they struggle to effectively handle occlusions and require extensive training and registration data.
Innovation Solution
An image recognition apparatus that determines an occlusion pattern in a facial image by comparing it with a standard image, allowing for identification by excluding regions based on the occlusion pattern, thereby improving accuracy without increasing system costs or processing load.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If existing image recognition apparatuses use standard facial image comparison, then recognition process is simple, but recognition accuracy deteriorates when face is partially occluded with wearing articles
Solution Approach 1:
The invention extracts and removes occluded regions from the facial image before performing recognition. The determination unit identifies occluded regions by comparing the input facial image with the standard facial image, and the exclusion unit removes these identified occluded regions. This allows the recognition unit to perform accurate matching using only the non-occluded regions, thereby resolving the contradiction between maintaining simple recognition processes and achieving high accuracy despite occlusions.
Solution Approach 2:
The invention segments the facial image into occluded and non-occluded regions. The determination unit divides the facial image by identifying which regions are occluded through comparison with the standard image. This segmentation enables the system to selectively process only the relevant non-occluded regions for recognition, improving accuracy without requiring complex retraining or additional data.
2Measurement precision
If multiple pieces of facial information are registered for each user in different wearing states, then recognition accuracy improves, but device complexity and processing load increase
Solution Approach 1:
The invention enables the system to automatically determine and exclude occluded regions during the recognition process itself, without requiring pre-registration of multiple facial images in different wearing states. The determination unit autonomously identifies occluded regions by comparing the input image with the standard image, and the exclusion unit automatically removes them. This self-service approach achieves high recognition accuracy while keeping the system simple and avoiding the need for extensive training data and complex processing.
3Measurement precision
If occluded regions are excluded from identification process, then recognition accuracy improves, but loss of information increases
Solution Approach 1:
The invention converts the harmful effect of occlusion into a beneficial filtering process. By determining which regions are occluded through comparison with the standard facial image and selectively excluding only those occluded regions from the recognition process, the system eliminates the negative impact of occlusions while preserving all useful non-occluded facial information. This approach improves recognition accuracy without unnecessary loss of information, as only the harmful occluded portions are removed.
Data Source
AI summary
An image recognition apparatus, method and non-transitory computer-readable storage medium are disclosed. An image recognition apparatus may include a memory storing instructions, and at least one processor configured to process the instructions to, determine an occlusion pattern of a face in a facial image representing the face by comparing the facial image with a standard facial image representing a standard face, and identify the facial image using a region determined by the occlusion pattern in the facial image.