Segmented Image Processing for Motion Vector Accuracy Across Sizes
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image processing methods using machine learning models for motion vector estimation face accuracy issues when the input image size during estimation differs from the size used during training, particularly when the input size is larger, due to variations in receptive fields and training data sets, leading to decreased estimation accuracy.
Innovation Solution
The method involves dividing the input image into smaller, predetermined-sized areas and processing each area separately using a trained machine learning model, ensuring the input size during estimation matches or is smaller than a reference size determined by the training size, thereby maintaining high estimation accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Length of moving object
If the input image size during estimation is made larger to handle high-resolution images, then the model can process more detailed images, but the estimation accuracy decreases due to receptive field variations and training data limitations
Solution Approach 1:
The patent divides the input image into multiple smaller image sets (e.g., first image set, second image set, third image set) and processes each segment separately through the machine learning model. This segmentation allows the model to maintain accurate motion vector estimation for each local region while handling larger overall images, resolving the contradiction between processing high-resolution images and maintaining estimation accuracy.
Solution Approach 2:
The patent dynamically adjusts the image size parameter used for estimation based on the actual input image dimensions. When the input image size exceeds the training size, the system modifies the estimation parameters and processes the image in segmented portions. This parameter adaptation allows the model to maintain accuracy across varying image sizes without being limited by the fixed training data dimensions.
2Manufacturing precision
If the machine learning model is trained using a specific image size, then the model parameters and receptive fields are optimized for that size, but the model performance deteriorates when processing images of different sizes
Solution Approach 1:
The patent creates a universal image processing system that can handle multiple image sizes by dividing the input image into different sized segments and processing them through the trained model. The system maintains the original training optimization while gaining versatility to process images of various sizes, effectively making the model multi-functional across different input dimensions.
Solution Approach 2:
The patent introduces dynamic image segmentation that adapts to the input image size. The system dynamically determines how to divide the image into manageable sets based on the actual input dimensions, allowing the model to maintain its trained performance while adapting to varying image sizes. This dynamic adjustment resolves the rigidity between fixed training parameters and variable input requirements.
3Measurement precision
If the input image is divided into smaller areas and processed separately, then the estimation accuracy is maintained, but the processing time and computational complexity increase
Solution Approach 1:
The patent segments the input image into multiple smaller image sets and processes them in parallel through the machine learning model. By dividing the processing task into independent segments that can be handled simultaneously, the system maintains high estimation accuracy for each region while reducing the overall processing time compared to processing the entire large image sequentially.
Data Source
AI summary
An image processing method includes acquiring, based on a first image set including a first image and a second image of a first size, a second image set of a second size smaller than the first size, which corresponds to partial areas of the first image set, and acquiring a motion vector by inputting the second image set into a machine learning model. The motion vector is a motion vector in the second image based on the first image. The machine learning model is trained using a third image set of a third size. The second size is equal to or smaller than a fourth size. The fourth size is set based on the third size.