Phase Detection Trimap Generation for Accurate Subject Separation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for generating Trimaps, such as Japanese Patent Laid-Open No. 2010-066802, lack accuracy when subjects and backgrounds have similar colors due to the absence of distance information.
Innovation Solution
Utilize an image plane phase detection sensor to obtain distance information, classifying image regions into foreground, background, and unknown regions based on distance distribution from multiple parallax images.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If distance information is not used (conventional method), then the device complexity is reduced, but the Trimap accuracy deteriorates when subject and background have similar colors
Solution Approach 1:
The patent introduces distance information as an intermediary element to resolve the ambiguity in color-based segmentation. By using depth data from the image plane phase detection sensor, the system can distinguish between foreground and background regions even when their colors are similar, thereby improving Trimap accuracy without requiring complex color analysis algorithms
Solution Approach 2:
The patent replaces conventional color-based image processing methods with a depth-based approach using image plane phase detection. This substitution leverages the optical phase detection mechanism to obtain distance information, which then serves as the primary basis for region classification, avoiding the limitations of color-only methods
2Measurement precision
If distance information from image plane phase detection sensor is used, then the Trimap accuracy is improved, but the device complexity increases due to additional sensor requirements
Solution Approach 1:
The patent achieves multi-functionality by utilizing the image plane phase detection sensor for both autofocus operation and distance information acquisition for Trimap generation. The same sensor hardware serves dual purposes: maintaining focus control while providing depth data for image processing, thereby avoiding the need for separate depth sensing devices
Solution Approach 2:
The image plane phase detection sensor performs self-service by providing its captured distance information directly to the Trimap generation process. The sensor's inherent phase detection capability automatically yields depth data that can be processed alongside the captured image, eliminating the need for additional active sensing components
3Manufacturing precision
If binary image is generated without distance information, then the processing speed is maintained, but the background separation accuracy deteriorates at boundaries
Solution Approach 1:
The patent applies local quality by using distance information specifically at boundary regions where foreground and background meet. The system classifies pixels into foreground, background, and unknown regions based on distance thresholds, providing enhanced accuracy precisely where it is most needed at the boundaries while maintaining efficient processing
Applied Scientific Principles
This section explains which scientific principles are used to turn an abstract innovation direction into a practical engineering solution.
Function Achieved in This Case
Generates a highly-accurate Trimap by distinguishing regions with precise distance-based classification, improving the accuracy of subject and background separation.
Implementation Method 1
a plurality of photoelectric conversion units are arranged, each photoelectric conversion unit receiving a light flux passing through a different partial pupil region of an imaging optical system
Data Source
AI summary
A generation unit generates a background separation image in which regions of a captured image are classified as a foreground region, a background region, and an unknown region, based on distance distribution information obtained from a plurality of parallax images. An output unit outputs the captured image and the background separation image. A region in which a distance in the distance distribution information is within a first range is classified as the foreground region. A region in which a distance in the distance distribution information is outside a second range broader than the first range is classified as the background region. A region in which a distance in the distance distribution information is outside the first range and inside the second range is classified as the unknown region.