Automatic Image Orientation Detection Using Face Analysis
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing digital image management software lacks an efficient method to automatically determine the correct orientation of images without EXIF orientation flags, requiring manual rotation by users.
Innovation Solution
A method that combines EXIF orientation tag analysis, face detection, and a classifier to determine the proper orientation of digital images by rotating images +90 and −90 degrees and applying a multi-class classifier when necessary, utilizing face detection algorithms and supervised learning.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If EXIF orientation tags are used to determine image orientation, then orientation information can be obtained, but the method fails when EXIF tags are missing or inaccurate
Solution Approach 1:
The patent introduces face detection as an intermediary method to determine image orientation. When EXIF orientation tags are missing or unreliable, the system uses face detection algorithms to identify the correct orientation by detecting faces in different rotated versions of the image, thereby bridging the gap between EXIF dependency and robust orientation determination
Solution Approach 2:
The system performs self-service orientation detection by automatically analyzing the image content itself (through face detection and classifier) rather than relying entirely on external metadata. The image is processed through multiple rotations and face detection to self-determine its correct orientation, making the system independent of EXIF tag availability
2Ease of operation
If manual rotation by users is required, then simplicity is maintained, but user convenience and time efficiency deteriorate
Solution Approach 1:
The system automatically performs orientation determination without requiring user intervention. By implementing face detection and classifier-based orientation identification, the software self-corrects image orientation, eliminating the need for users to manually rotate images and saving time while improving convenience
3Measurement precision
If multiple rotation operations are performed to detect orientation, then orientation accuracy improves, but processing time and computational complexity increase
Solution Approach 1:
The patent applies partial action by rotating the image only in specific directions (+90 and -90 degrees) rather than all possible angles. This limited set of rotations is sufficient for orientation detection since images are typically oriented in these discrete directions, achieving accurate orientation identification while minimizing unnecessary computational operations
Solution Approach 2:
The system performs preliminary rotation operations to generate multiple oriented versions of the image before final orientation determination. By pre-rotating the image +90 and -90 degrees and detecting faces in these versions, the system prepares the data needed for accurate orientation identification without requiring complex real-time processing
Data Source
AI summary
A method determining image orientation includes determining if an image includes an orientation tag and if the tag indicates the image is rotated +90 or −90 degrees. When the image does not include the tag or the tag does not indicate the image is rotated, the method includes determining if a face is detected in the original image and displaying the original image when a face is detected. When a face is not detected, the method includes rotating the image +90 and −90 degrees and detecting a face in the rotated images. When a face is not detected, the method includes applying a classifier to determine the image's proper orientation. When a face is detected in one rotated image, the method includes displaying the rotated image. When a face is detected in both rotated images, the method includes applying the classifier to determine the image's proper orientation.