Image Colorization via Intrinsic Distance Propagation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image processing techniques require significant computational resources and often produce artifacts like block distortion when colorizing monochrome images or videos, and they struggle with recoloring images that already contain color information, necessitating complex and slow methods such as optical flow for video processing.
Innovation Solution
The method employs user-supplied 'scribbles' to propagate color changes based on intrinsic distances within the image, using geodesic distances calculated from luminance gradients, allowing for fast and accurate color manipulation by blending chrominance values weighted by their proximity to user-defined colors, which can be applied to still images and videos.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If optical flow is used for video colorization, then color manipulation can be performed on video sequences, but the processing time becomes excessively long and computational complexity increases
Solution Approach 1:
The patent extracts and removes the optical flow computation step from the video colorization pipeline. Instead of using optical flow to track pixel movements between frames, the invention directly propagates color information from scribbled regions across the video sequence using a simplified distance transform approach, eliminating the computationally expensive optical flow calculation while maintaining color consistency across frames
Solution Approach 2:
The patent performs preliminary color propagation from user-scribbled regions to generate a complete color map for each frame before final composition. This preliminary action computes intrinsic distances from scribble regions to all other pixels and propagates colors along gradient paths in advance, allowing fast rendering without requiring complex per-pixel optimization during playback or interaction
2Productivity
If propagation is performed in all directions from seed pixels, then colorization speed is improved, but visible block distortion artifacts appear
Solution Approach 1:
The patent segments the color propagation process into two distinct phases: (1) computing intrinsic distances from scribble regions to all pixels using gradient-weighted paths, and (2) propagating colors along these pre-computed gradient paths. This segmentation allows fast directional propagation along image gradients while maintaining color blending accuracy, avoiding the block artifacts caused by uniform all-directional propagation
Solution Approach 2:
The patent applies different propagation strategies in different regions of the image based on local gradient characteristics. By propagating colors along gradient paths rather than in all directions uniformly, the method adapts to local image structure, maintaining high color blending accuracy in regions with strong gradients while still achieving fast processing overall
3Measurement precision
If manual segmentation and Bayesian image matting are used, then accurate alpha channel estimation is achieved, but computational cost becomes intensive
Solution Approach 1:
The patent replaces the computationally expensive manual segmentation and Bayesian matting pipeline with a much simpler and faster approach: computing intrinsic distances from scribble regions using gradient-weighted distance transforms and propagating colors along these paths. This disposable, lightweight method achieves sufficient alpha channel accuracy for colorization without the intensive computational cost of iterative Bayesian optimization
Data Source
AI summary
Images are colorized, recolorized, or otherwise manipulated by adding scribbles of a blending medium (such as chrominance) to a measurement medium (such as luminance) of the image. The blending medium propagates from the scribbles according to an intrinsic or geodesic distance that is a function of the measurement medium. Intrinsic distances may be weighted. An effect scribble may produce an effect, such as result data, in the blending medium according to a predefined function. A preprocessing scribble may be converted to one or more effect scribbles which are then propagated. A measurement scribble may propagate so as to alter the intrinsic distances of other scribbles.