Image Processing Apparatus Correcting Target Region Distortion
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Typical image processing methods face inefficiencies due to the stretching of objects during entire image correction, leading to poor processing effects and slow speeds, especially when dealing with distorted images like fisheye images.
Innovation Solution
The method involves generating feature maps from scaled images using a deformable convolutional neural network (CNN), detecting target regions, and spatially transforming these regions to prevent distortion, allowing for accurate feature extraction and improved object processing without correcting the entire image.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Shape
If the entire fisheye image is corrected before object processing, then the image distortion is removed, but the object becomes greatly stretched leading to poor processing effects
Solution Approach 1:
The patent divides the image processing into two stages: first correcting the entire image geometry, then detecting target regions and applying localized inverse transformation only to those regions. This segmentation allows the global distortion to be corrected while preserving local object shapes through selective region-based processing.
Solution Approach 2:
The patent applies different processing qualities to different parts of the image: the entire image undergoes geometric correction, but only detected target regions undergo additional inverse transformation to restore local object shapes. This local quality adjustment prevents object stretching while maintaining overall image correction.
2Shape
If the entire image is corrected before object processing, then the image distortion is removed, but the image processing speed becomes slow
Solution Approach 1:
The patent segments the image correction process by first applying a lightweight global correction to the entire image, then detecting target regions, and finally applying computationally intensive inverse transformation only to those specific regions. This segmentation reduces overall computational load and improves processing speed.
Solution Approach 2:
Instead of applying full inverse transformation to the entire image, the patent applies the transformation partially only to detected target regions. This partial action significantly reduces computational complexity while maintaining the benefits of distortion correction for objects of interest.
3Productivity
If features are extracted from distorted images without correction, then processing speed is maintained, but sampling distortion occurs reducing accuracy
Solution Approach 1:
The patent performs preliminary global geometric correction on the entire image before feature extraction and target detection. This preliminary action removes the severe distortion that would otherwise cause sampling errors, ensuring accurate feature extraction while maintaining processing efficiency through the subsequent selective regional processing.
Data Source
AI summary
A processor-implement method with image processing includes: generating a feature map of a first image and detecting a target region in the first image based on the feature map; correcting the detected target region; and processing an object corresponding to the target region, based on the corrected target region.