Multi-Scale Image Saliency Processing for Automatic Zoom Visualization
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current image processing technologies lack an effective method to automatically determine and display salient regions of an image at various spatial scales, leading to incomplete or irrelevant visualizations during image zooming and viewing.
Innovation Solution
A method that generates saliency data by determining image features, processing it with weighting functions at multiple spatial scales to identify regions of interest, and producing response data to select salient image portions, which are then used to create a visualization path that traverses these regions based on scale and relative distance.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If automatic zoom is performed using predetermined magnification factors or simple saliency detection, then the operation is simple and fast, but the visualization may miss important large-scale structures or fail to provide a comprehensive visual tour of the image
Solution Approach 1:
The patent introduces a scale dimension by processing saliency data at multiple spatial scales simultaneously. Instead of analyzing the image at a single scale, the system convolves saliency data with Gaussian kernels of different standard deviations to create scale-space representation, allowing detection of salient regions at various sizes and ensuring comprehensive coverage from large-scale structures to fine details
Solution Approach 2:
The patent segments the saliency analysis process by separating detection at different scales. Each scale independently identifies salient regions appropriate to its resolution level, and these segmented results are then combined through max pooling to create a comprehensive visualization path that includes both large-scale and fine-scale important regions
2Measurement precision
If the system focuses on detecting fine details, then small features are captured, but large-scale structures are neglected
Solution Approach 1:
The patent adds the scale dimension to the analysis by processing the same saliency data at multiple spatial resolutions simultaneously. Each scale level captures features appropriate to its resolution, with finer scales detecting detailed structures and coarser scales detecting large-scale patterns, all within a unified scale-space framework
Solution Approach 2:
The patent changes the spatial scale parameter systematically by varying the standard deviation of Gaussian kernels used for convolution. This parameter change allows the same saliency detection algorithm to operate effectively at different resolutions, capturing both fine details and large-scale structures through controlled modification of the scale parameter
3Loss of information
If multiple scales are processed to capture all image features, then comprehensive visualization is achieved, but computational complexity increases
Solution Approach 1:
The patent performs preliminary saliency detection at a coarse scale first to identify major regions of interest. These pre-identified regions then guide the subsequent fine-scale analysis, allowing the system to focus computational resources on areas most likely to contain salient features rather than uniformly processing the entire image at all scales
Solution Approach 2:
The patent applies a limited number of discrete scales (typically 3-5 Gaussian kernel sizes) rather than continuous scale processing. This partial action approach provides sufficient multi-scale coverage to capture both large-scale structures and fine details while keeping computational complexity manageable through selective sampling of scale levels
Data Source
AI summary
In a method of and an apparatus for processing image data representing an image saliency data for the image is generated by determining a series of features of the image and the determined features are used to generate a probability measure for each point of the image representative of a location of a subject of the image. The saliency data is processed using respective ones of weighting functions of a plurality of spatial scales in order to determine the positions of regions of interest of the image at respective ones of the scales. Response data is generated for each scale representing the relative strength of response of the saliency data at the positions of the determined regions of interest to the function at that scale.