Image Processing Device for Accurate Important Area Extraction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional image processing technologies face difficulties in extracting important areas from images while considering the actual capturing direction, particularly with wide-angle or full-spherical cameras, where lens distortion and angular differences between central and peripheral areas lead to inaccurate important area selection.
Innovation Solution
An image processing device with an importance calculating unit, capturing-direction acquiring unit, and parameter estimating unit that calculates importance and captures directions in a three-dimensional space, treating importance distributions as a mixture distribution of element distributions to estimate parameters and set important areas accordingly, using methods like von Mises-Fisher distribution and EM algorithm.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If clustering is conducted on the basis of positions within the image, then important areas can be extracted, but the relation of the actual capturing directions cannot be taken into consideration, leading to inaccurate important area selection in wide-angle or full-spherical images
Solution Approach 1:
The patent transforms the problem from two-dimensional image space clustering to three-dimensional capturing direction space clustering. By converting image positions to capturing directions (azimuth and elevation angles) and performing clustering in this 3D spherical coordinate system, the method accurately reflects the actual angular relationships between pixels, especially for wide-angle and full-spherical cameras where peripheral pixels have different angular densities compared to central pixels.
2Manufacturing precision
If lens distortion is corrected by using the pinhole camera model, then distortion is reduced, but an angular difference remains between the shooting angle between neighboring pixels in the image central area and the shooting angle between neighboring pixels in the image peripheral area
Solution Approach 1:
The patent changes the parameter space from Cartesian image coordinates to spherical angular coordinates (azimuth and elevation). Instead of working with pixel positions (x, y) that suffer from distorted angular relationships after lens correction, the method transforms to angular parameters (θ, φ) that directly represent the capturing directions. This parameter transformation preserves the true angular relationships between rays regardless of their position in the image.
3Productivity
If conventional clustering methods are used on importance distribution, then important areas can be identified, but it is difficult to set important areas while considering the influence of the actual capturing direction
Solution Approach 1:
The patent introduces capturing direction (azimuth and elevation angles) as an intermediary between image position and importance. Instead of directly clustering importance values at image positions, the method first maps positions to capturing directions, then performs clustering in this intermediate angular space. This intermediary transformation ensures that the clustering respects the actual geometric relationships of the camera's field of view, improving both accuracy and efficiency.
Data Source
AI summary
An image processing device includes an importance calculating unit, a capturing-direction acquiring unit, a parameter estimating unit, and an important-area setting unit. The importance calculating unit is configured to calculate importance of each of a plurality of positions in an image. The capturing-direction acquiring unit is configured to acquire, for each of the positions, a capturing direction in a three-dimensional space. The parameter estimating unit is configured to regard importance distributions in respective capturing directions as a mixture distribution made up of element distributions, so as to estimate a parameter of each of the element distributions. The important-area setting unit is configured to set an important area from the image in accordance with the parameter.