Combined Image and Depth Kernels for Halo-Free Bokeh
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image and video modification techniques, such as the artificial-bokeh technique, often result in halo effects due to color mixing at depth discontinuities, particularly when blurring foreground and background pixels.
Innovation Solution
A method involving the generation of image-based and depth-based kernels, combined to modify pixels based on blur radii derived from depth information, to minimize halo effects by selectively blurring background pixels while preserving foreground pixels.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If artificial-bokeh technique is used to blur background pixels, then depth separation effect is improved, but halo effects and color mixing occur at depth discontinuities
Solution Approach 1:
The patent applies different kernel characteristics to different regions of the image based on depth information. For each pixel, a kernel is selected and adjusted according to the depth discontinuities in its local neighborhood, allowing selective blurring that adapts to local scene structure. This resolves the contradiction by making the blurring operation locally optimized rather than applying a uniform artificial-bokeh effect throughout the image.
Solution Approach 2:
The patent dynamically adjusts kernel parameters (such as blur radius and spread) based on depth information and scene analysis. The kernels are not fixed but adapt their characteristics according to the detected depth discontinuities and local image content, enabling the system to respond to varying scene conditions and minimize halo effects where depth changes occur.
2Reliability
If pixels are blurred based on depth information, then foreground-background separation is improved, but unwanted color blending occurs at depth boundaries
Solution Approach 1:
The patent introduces depth information as an intermediary factor that mediates between the image pixels and the blurring operation. Instead of directly blurring pixels based solely on color or intensity, the depth data serves as a guiding intermediary that determines which pixels should be blurred and to what extent, thereby achieving reliable separation while preventing unwanted color blending at depth boundaries.
Solution Approach 2:
The patent changes the parameters of the blurring kernel dynamically based on depth information. By adjusting kernel characteristics (blur radius, spread, shape) according to depth discontinuities and local scene properties, the system optimizes the blurring process to maintain reliable foreground-background separation while minimizing harmful color blending effects at depth boundaries.
Data Source
AI summary
Systems and techniques are described herein for modifying image data. For instance, a method for modifying an image is provided. The method may include determining a first kernel for a pixel of an image based on pixel values of a window of pixels of the image; determining a second kernel for the pixel of the image based on depth data of a window of depth information, wherein the depth information is related to the image; combining the first kernel with the second kernel to generate a combined kernel; and modifying the pixel based on the combined kernel.