Contour Shape Recognition via Multi-Scale Feature Maps
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing contour shape recognition methods face challenges in accurately representing and classifying shape features due to incomplete shape information, sensitivity to local changes, and high computational complexity.
Innovation Solution
A contour shape recognition method that involves sampling and extracting salient feature points, calculating shape feature functions using three types of shape descriptors, and synthesizing these features into a color representation image for input into a two-stream convolutional neural network for classification.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If manual shape descriptors are used to extract contour features, then the extraction process is simple, but the shape information is incomplete and sensitive to local changes
Solution Approach 1:
The patent segments the contour shape into multiple sampling points and extracts features at different scales (semi-global and full-scale). By dividing the contour into discrete points and analyzing them at multiple resolution levels, the method captures comprehensive shape information while maintaining systematic processing
Solution Approach 2:
The patent introduces a scale dimension by calculating shape feature functions at both semi-global scale and full-scale space. This multi-scale approach transforms the 2D contour problem into a 3D feature space (x, y, scale), enabling the system to capture shape characteristics that are invisible at single scale while remaining computationally tractable
2Loss of information
If multiple shape descriptors are added to improve shape representation, then the shape information becomes more complete, but the computational complexity increases
Solution Approach 1:
The patent merges three different shape descriptors (area, arc length, and barycentric distance) into a unified shape feature function framework. By combining these descriptors mathematically rather than processing them separately, the method achieves comprehensive shape representation while reducing redundant computations
Solution Approach 2:
The patent applies partial action by selectively computing shape features only at critical sampling points rather than all contour points. The adaptive sampling strategy computes descriptors at representative locations, providing sufficient shape information without the computational burden of processing every point
3Measurement precision
If convolutional neural network is directly applied to contour shape recognition, then the recognition capability is enhanced, but the recognition effect is low due to lack of surface texture and color information
Solution Approach 1:
The patent introduces shape feature grayscale maps as an intermediary representation between the raw contour and the CNN classifier. These maps encode shape information in a format suitable for CNN processing, acting as a bridge that translates geometric contour data into a representation that leverages the power of deep learning while compensating for the lack of texture and color
Solution Approach 2:
The patent transforms the contour shape into multiple parameter representations including area, arc length, and barycentric distance at different scales. This parameter transformation converts the raw geometric data into a feature space that preserves shape characteristics while being amenable to CNN-based classification
4Measurement precision
If adaptive sampling is used to extract salient feature points, then the shape features are more accurate, but the processing time increases
Solution Approach 1:
The patent performs preliminary action by pre-processing the contour to identify salient feature points before the main recognition task. By detecting key geometric features (such as corners, inflection points, and extreme points) in advance, the system reduces the number of points requiring detailed analysis, thereby improving accuracy while controlling processing time
Data Source
AI summary
Provided is a contour shape recognition method, including: sampling and extracting salient feature points of a contour of a shape sample; calculating a feature function of the shape sample at a semi-global scale by using three types of shape descriptors; dividing the scale with a single pixel as a spacing to acquire a shape feature function in a full-scale space; storing feature function values at various scales into a matrix to acquire three types of feature grayscale map representations of the shape sample in the full-scale space; synthesizing the three types of grayscale map representations of the shape sample, as three channels of RGB, into a color feature representation image; constructing a two-stream convolutional neural network by taking the shape sample and the feature representation image as inputs at the same time; and training the two-stream convolutional neural network, and inputting a test sample into a trained network model to achieve shape classification.