Automated Collage Assembly Using Saliency Extraction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for automating collage creation from images fail to effectively eliminate non-salient regions, resulting in collages that do not accurately represent the artistic intent of extracting expressive regions of interest, and are inefficient in terms of space and aesthetics.
Innovation Solution
A computerized method for forming collages by detecting and extracting salient regions using a dissimilarity measure that combines appearance and distance components, and assembling these regions in a puzzle-like manner to maximize saliency, compactness, and minimize occlusion, allowing for non-rectangular shapes and incorporating high-level features like face recognition.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Extent of automation
If rectangular salient regions are extracted and assembled using existing automated methods, then the collage construction is simplified and automated, but non-salient regions cannot be effectively eliminated and the artistic intent is not accurately represented
Solution Approach 1:
The patent extracts only the salient regions from each photograph by comparing pixel dissimilarity to background, eliminating non-salient regions. This is achieved by computing a dissimilarity measure for each pixel and selecting only those pixels that exceed a threshold, thereby taking out the essential content while discarding redundant background information.
Solution Approach 2:
The patent applies different processing to different regions of the image based on their saliency. Salient regions are extracted and preserved with high fidelity, while non-salient regions are eliminated. This local differentiation ensures that important content is maintained while unnecessary background is removed, accurately representing artistic intent.
2Ease of manufacture
If rectangular regions are used for collage assembly, then the assembly process is simpler, but the aesthetic appeal and space efficiency are reduced
Solution Approach 1:
The patent employs rounded corner rectangles instead of sharp rectangular shapes for the collage pieces. This curvature modification enhances the aesthetic appeal by creating softer, more visually pleasing transitions between pieces, while still maintaining the structural simplicity of rectangular-based assembly. The rounded corners improve compactness and visual harmony without significantly complicating the assembly process.
3Loss of information
If salient regions are extracted using dissimilarity measures, then non-salient regions are eliminated, but the computational complexity increases
Solution Approach 1:
The patent computes dissimilarity measures for all pixels to ensure complete elimination of non-salient regions, but then applies a threshold to select only the most salient pixels. This partial action approach ensures thorough saliency detection while avoiding the need to process every pixel with equal computational resources, as only pixels exceeding the threshold are retained for the final collage.
Data Source
AI summary
A computerized method of image processing to form a collage within a predetermined outline from a plurality of images, the method comprising: processing each image to assign a saliency measure to each pixel, said processing utilizing a dissimilarity measure which combines an appearance component and a distance component; finding a first patch of said image and comparing with other patches at different distances from said first patch using said dissimilarity measure, thereby to obtain a score; applying to each pixel of said first patch said obtained score; continuing said comparing and scoring with additional patches of said image until each pixel obtains a score; from said scored pixels providing for each image a region of interest, by setting an initial boundary that encloses a predetermined set of highest scored pixels, and propagating a curve around said initial boundary in such a way as to minimize length and maximize included saliency; and combining said regions of interest into said collage by: ordering said image regions by importance; placing successive regions within said predetermined outline, so as to maximize saliency and compactness and minimize occlusion.