Image Processing Apparatus Feature Point Extraction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing techniques for calculating motion vectors in video images face accuracy degradation due to blur caused by photographic system characteristics and position shifts of feature points, particularly when extracting features from images other than the original size.
Innovation Solution
An image processing apparatus that reduces images to a predetermined ratio, specifies feature points in the reduced image, designates corresponding points in the original image using a filter, and calculates motion vectors based on these points to improve accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Speed
If feature points are extracted from reduced images to improve extraction speed, then processing speed is improved, but feature point position accuracy deteriorates due to blur and position shift
Solution Approach 1:
The patent divides the image processing into multiple stages: first extracting feature points from reduced images for speed, then refining their positions by searching in corresponding regions of original images. This segmentation allows the system to benefit from both fast extraction and high accuracy without compromising either aspect.
Solution Approach 2:
The patent performs preliminary feature point extraction from reduced images before processing original images. This preliminary action identifies candidate feature points quickly, and then the system refines these points by searching in corresponding regions of original images, ensuring both speed and accuracy are achieved.
2Productivity
If feature points are extracted from images other than reference images, then motion vector calculation can be performed, but calculation accuracy deteriorates due to blur influence
Solution Approach 1:
The patent uses reduced images as an intermediary to extract feature points, which then serve as references for finding corresponding points in original images. This intermediary approach allows the system to work with multiple images while maintaining accuracy by using the reduced image feature points to guide the search in original images.
Solution Approach 2:
The patent creates a correspondence relationship between feature points in reduced images and their corresponding regions in original images. By copying the feature point identification from reduced images to original images through this correspondence relationship, the system maintains calculation accuracy while being able to process multiple images.
3Measurement precision
If the reference image is used for feature point extraction, then position accuracy is maintained, but processing speed decreases
Solution Approach 1:
The patent segments the feature point extraction process into two parts: extracting feature points from reduced images for speed, and then refining them using original images for accuracy. This segmentation allows the system to achieve both high speed and high accuracy without having to process only original images.
Solution Approach 2:
The patent changes the image resolution parameter by using reduced images for initial feature point extraction. This parameter change allows faster processing while the system later refines the results using original images, achieving both speed and accuracy requirements.
Data Source
AI summary
A processor executes a program and reduces reference and comparative images to reduced reference and comparative images, in a ratio of ½, by setting a pixel signal value of each pixel in each image using pixel signal values for each of four pixels in each of the images. The processor specifies (i) a first corresponding point similar to a first feature point included in the reduced reference image, from the reduced comparative image, using a first filter for 3×3 pixels, and (ii) a second corresponding point similar to a second feature point in the reference image, corresponding to the first feature point, from the comparative image, using a second filter for 5×5 pixels, on the basis of a position of the first corresponding point specified in the reduced comparative image.