Image Resizing via Significance-Guided Spline Deformation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image resizing methods, such as seam carving and discrete transformation techniques, suffer from artifacts due to discontinuities and fold-overs, which affect the quality and compressibility of resized images.
Innovation Solution
A computer-implemented method that computes image significance, uses a diffeomorphic transformation parameterized as an interpolating spline with control points to resize images, avoiding fold-overs and artifacts by deforming cells with lower weighted averages, and applying the transformation to achieve a resized image with improved smoothness and compressibility.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Volume of moving object
If seam carving is used to resize images by removing lines of adjacent pixels, then the image size is reduced, but artifacts are produced due to discontinuities in the image
Solution Approach 1:
The patent replaces the discrete mechanical removal of pixel lines with a continuous mathematical transformation field. Instead of sequentially removing adjacent pixel lines, the invention uses a continuous transformation function that smoothly remaps all pixel positions, eliminating the discontinuities that cause artifacts while achieving the same resizing effect.
Solution Approach 2:
The patent introduces a continuous transformation field as an intermediary between the original and resized images. This transformation field acts as a mediator that smoothly transitions pixel positions, avoiding direct discrete removal operations and their associated artifacts.
2Volume of moving object
If discrete transformation methods are used to remove or aggregate pixels, then the image size is changed, but serious artifacts are generated due to discontinuities
Solution Approach 1:
The patent replaces discrete transformation operations with a continuous transformation field. Instead of removing or aggregating pixels in discrete steps, the invention uses a continuous mathematical function that smoothly remaps all pixel positions simultaneously, eliminating the discontinuities inherent in discrete methods.
3Stability of the object's composition
If continuous transformation methods with interpolation are used to resize images, then the transformation is smooth, but artifacts are created on the edges of cells due to piecewise smooth transformation
Solution Approach 1:
The patent segments the transformation into a grid of control points that define a continuous transformation field. By using control points arranged in a regular grid pattern, the transformation achieves global smoothness while avoiding the piecewise discontinuities that occur at cell boundaries in traditional methods.
Solution Approach 2:
The patent changes the parameterization of the transformation from piecewise linear (affine) to a continuous field defined by control points. This parameter change enables smooth transitions across the entire image domain while maintaining computational efficiency through the structured control point grid.
4Stability of the object's composition
If continuous transformation methods are used, then pixel positions are smoothly adjusted, but fold overs occur where pixels disappear from the image
Solution Approach 1:
The patent incorporates a feedback mechanism through the significance map that guides the transformation. The significance map provides feedback on which regions are important, constraining the transformation to preserve those regions while allowing flexibility in less significant areas, thus preventing fold overs and pixel loss.
Solution Approach 2:
The patent changes the transformation parameters locally based on the significance map. By adjusting the transformation field according to regional importance, the method ensures that significant regions maintain their pixel structure while less significant regions can be smoothly transformed, preventing fold overs.
Data Source
Figure 1
Figure 2
Figure 3~6
AI summary
The invention notably relates to computer-implemented method for resizing an image I. The method comprises the steps of: - providing the image I to resize; - providing an image significance by computing a significance of each pixel in the image to resize; - extracting an original spatial domain of the significance image; - providing a transformation Tθ, parameterized as an interpolating spline by a set of control points, from Ω to a resized spatial domain Ω'; - subdividing Ω into cells, each cell being defined by a subset of control points of the set; - computing, for each cell, a weighted average of the significance of the pixels in the cell, the weighted average being computed using the interpolating spline; - deforming the cells of Ω by displacing one or more control points of the cells having a lesser weighted average; - computing the transformation Tθ over a spatial domain of the image I using the displaced one or more control points; and - computing a resized image J by applying the computed transformation Tθ to the spatial domain of the image I to resize.