Super Pixel Grouping for Object Recognition
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current machine learning and image processing technologies face challenges in effectively recognizing objects in images, particularly in earth-like environments, due to difficulties in accurately grouping pixels and distinguishing between different objects under varying lighting conditions.
Innovation Solution
The method involves classifying pixels into super pixels using a color space like CIELAB, which models human vision perception, by calculating combined distances based on color and geographic proximity, and applying weighting factors to optimize grouping, especially under different lighting conditions. This process is enhanced by dividing computation tasks across multiple processor cores for real-time applications.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If pixels are grouped into super pixels using color distance calculation, then object recognition accuracy is improved, but computation complexity increases
Solution Approach 1:
The patent segments the image processing task into two stages: first grouping pixels into super pixels based on color and spatial proximity, then performing object recognition on the super pixels. This segmentation reduces the number of processing units from millions of pixels to thousands of super pixels, thereby reducing computation complexity while maintaining recognition accuracy.
Solution Approach 2:
The patent merges multiple pixels into super pixels by combining color information (using CIELAB color space) and spatial proximity information. This merging process groups pixels with similar characteristics into unified super pixels, reducing the overall data volume for subsequent recognition tasks while preserving important visual features.
2Measurement precision
If color distance calculation is performed between seed pixels and proximal pixels, then super pixel grouping accuracy is improved, but processing time increases
Solution Approach 1:
The patent performs preliminary actions by first selecting seed pixels and calculating their color distances to proximal pixels before final super pixel formation. This preliminary grouping based on color proximity allows for efficient organization of pixels into candidate super pixels, reducing the search space for subsequent recognition steps.
Solution Approach 2:
The patent applies local quality by using CIELAB color space which models human vision perception, making the color distance calculation more aligned with perceived color differences. This localized improvement in distance metric accuracy enhances grouping precision without requiring global recalibration of all pixel relationships.
3Productivity
If multiple processor cores are used for computation, then processing speed is improved, but device complexity increases
Solution Approach 1:
The patent segments the computation task across multiple processor cores, with each core handling specific super pixel processing operations. This parallelization divides the workload into independent units that can be processed simultaneously, achieving speedup proportional to the number of cores while maintaining manageable complexity through task division.
Data Source
AI summary
A method is provided for recognition of a sky portion, a vertical object portion and a ground portion in an image. The image into a plurality of pixel sets by the electronic system. Expected values of each pixel sets with a sky distribution function, a vertical object distribution function and a ground distribution function by the electronic system are calculated and compared for each pixel set for determine each pixel set belonging to one of the sky portion, the vertical object portion or the ground portion.