Image Seam Determination for Foreground-Background Separation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image combining technologies, such as those using graph cut techniques, face challenges in accurately determining the seam between two photographed images, leading to inconveniences in combining images of people without overlap and requiring complex calculations, which results in low detection accuracy and inefficient processing.
Innovation Solution
An image processing apparatus and method that determines a seam candidate region by removing foreground objects, calculates differences between images, and uses a threshold value to identify suitable seams, allowing for simpler image combination by avoiding human regions and using a center-penetrating region when necessary.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If graph cut techniques are used to adaptively determine seam positions, then the seam can avoid human regions, but the detection accuracy remains low and the computational complexity increases
Solution Approach 1:
The patent segments the image processing task into distinct components: foreground object detection, seam candidate region determination, and seam determination. By dividing the complex graph cut process into manageable segments, the system achieves accurate seam positioning while reducing overall computational complexity.
Solution Approach 2:
The patent performs preliminary foreground object detection and seam candidate region determination before final seam calculation. This preliminary action identifies regions where seams should not pass, constraining the search space and reducing the computational burden of the subsequent graph cut optimization.
2Measurement precision
If graph cut techniques compute smoothing parameter and data parameter to minimize total energy function, then seam positioning becomes adaptive, but the processing becomes complex and inefficient
Solution Approach 1:
The patent applies different processing strategies to different regions of the image. In foreground regions, simple object detection suffices, while in background regions, the full graph cut optimization with smoothing and data parameters is applied. This localized quality approach maintains precision where needed while improving overall processing efficiency.
Solution Approach 2:
The patent dynamically adjusts parameter values based on regional characteristics. The smoothing parameter and data parameter are modified according to whether a region contains foreground objects or background, allowing adaptive seam positioning without uniformly applying complex calculations across the entire image.
3Device complexity
If a straight line seam is used for image combining, then the combining process is simple, but it requires photographing people so they do not overlap the line which is inconvenient
Solution Approach 1:
The patent transitions from a static straight line seam to a dynamic adaptive seam that adjusts its position based on detected foreground objects. The seam can bend and shift to avoid human regions while maintaining a relatively simple combining process, thus resolving the contradiction between simplicity and convenience.
Data Source
AI summary
There is provided an image processing apparatus that combines a first photographed image including a first foreground object and a background with a second photographed image including a second foreground object and the background, and generates an output image including the first foreground object, the second foreground object, and the background, the image processing apparatus including a seam candidate region determining unit that determines a seam candidate region obtained by removing regions of the first foreground object and the second foreground object from the first photographed image or the second photographed image, a seam determining unit that determines a seam which passes through a top edge and a bottom edge of the seam candidate region and divides the seam candidate region, and a combining unit that combines the first photographed image and the second photographed image at the seam to generate the output image.