Edge Detection via Line Merging and Neural Network Region Identification
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current image processing techniques face challenges in accurately detecting object edges from images that contain both the object and its external environment, making it difficult to extract relevant information effectively.
Innovation Solution
An edge detection method that involves processing images to obtain a line drawing of grayscale contours, merging similar lines to form initial and target lines, determining reference boundary lines using a boundary matrix, and employing a pre-trained neural network-based boundary line region identification model to identify edge regions and determine the object's edges.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional edge detection algorithms are used on images containing objects and external environment, then the detection process can be performed, but the accuracy of object edge detection deteriorates due to interference from external environment elements
Solution Approach 1:
The patent segments the image processing task into multiple stages: initial edge detection to obtain candidate lines, merging of similar lines to form initial merged lines, identification of boundary line regions using neural network, and final determination of target boundary lines. This segmentation allows the system to progressively filter out external environment interference while maintaining object edge detection accuracy.
Solution Approach 2:
The patent introduces boundary line region identification models based on neural networks as an intermediary component. These models act as a mediator between the initial edge detection results and the final edge extraction, enabling the system to distinguish object boundaries from external environment elements by learning patterns from training data.
2Measurement precision
If multiple processing steps are applied to improve edge detection accuracy, then the detection precision improves, but the processing time and computational complexity increase
Solution Approach 1:
The patent performs preliminary merging of similar lines to form initial merged lines before boundary line region identification. This preliminary action reduces the number of candidate lines that need to be processed by the neural network model, thereby reducing computational complexity and processing time while maintaining detection accuracy.
Solution Approach 2:
The patent merges similar lines in the image to form initial merged lines, and subsequently merges initial merged lines that fall within the same boundary line region to form target boundary lines. This merging operation reduces the total number of lines to be processed, decreasing computational load and processing time while preserving essential edge information.
3Productivity
If simple line merging is performed without neural network identification, then the processing speed is maintained, but the ability to distinguish object boundaries from background lines deteriorates
Solution Approach 1:
The patent introduces boundary line region identification models based on neural networks as an intermediary component. These models act as a mediator between the initial edge detection results and the final edge extraction, enabling the system to distinguish object boundaries from external environment elements by learning patterns from training data.
Solution Approach 2:
The patent transforms the boundary line identification problem into a parameter classification task by training neural networks to recognize patterns in boundary line regions. The model learns to identify characteristic parameters of object boundaries versus background elements, significantly improving identification reliability while maintaining acceptable processing speeds through efficient model architecture.
Data Source
AI summary
The invention provides an edge detection method and a device of an object in an image, an electronic equipment, and a computer-readable storage medium. The method includes: a line drawing of a grayscale contour in the image is obtained; similar lines in the line drawing are merged to obtain initial merged lines, and a boundary matrix is determined according to the initial merged lines; similar lines in the initial merged lines are merged to obtain target lines, and unmerged initial merged lines are also used as target lines; reference boundary lines are determined from the target lines according to the boundary matrix; boundary line regions of the object in the image are obtained; a target boundary line corresponding to the boundary line region is determined from the reference boundary lines; an edge of the object in the image is determined according to the determined target boundary lines.