Two-Stage Feature Extraction for Object Detection Accuracy
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Object detection in images, particularly face detection, faces challenges due to variations in pose, expression, background, illumination, and skin color, leading to low robustness and accuracy in conventional methods.
Innovation Solution
A two-stage feature extraction approach is employed, where the first stage extracts features characterizing the whole image scale and the second stage focuses on local features within identified candidate object regions, using convolutional neural networks to improve detection accuracy by considering both context and fine-grained details.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional single-stage feature extraction is used, then device complexity is low, but object detection accuracy deteriorates due to inability to capture both global context and local details
Solution Approach 1:
The patent divides the feature extraction process into two distinct stages: a first stage that extracts global features from the entire image to identify candidate object regions, and a second stage that extracts local features from identified candidate regions. This segmentation allows the system to capture both global context and local details, thereby improving object detection accuracy while managing computational complexity through hierarchical processing.
2Reliability
If full image processing is performed to ensure no candidate regions are missed, then object detection reliability improves, but computation time increases significantly
Solution Approach 1:
The patent performs preliminary feature extraction at a reduced resolution in the first stage to identify candidate object regions before conducting detailed feature extraction in the second stage. This preliminary action filters out non-candidate regions early, ensuring that computationally intensive processing is applied only to regions likely to contain objects, thereby maintaining detection reliability while significantly reducing overall computation time.
3Measurement precision
If high-resolution feature extraction is applied to entire image, then local detail detection improves, but energy consumption increases
Solution Approach 1:
The patent applies high-resolution feature extraction only to identified candidate object regions in the second stage, while the first stage operates at lower resolution across the entire image. This local quality approach ensures that computationally intensive and energy-consuming high-resolution processing is applied only where necessary, maintaining local feature detection precision while significantly reducing overall energy consumption compared to processing the entire image at high resolution.
Data Source
AI summary
In implementations of the subject matter described herein, a solution for object detection is proposed. First, a feature(s) is extracted from an image and used to identify a candidate object region in the image. Then another feature(s) is extracted from the identified candidate object region. Based on the features extracted in these two stages, a target object region in the image and a confidence for the target object region are determined. In this way, the features that characterize the image from the whole scale and a local scale are both taken into consideration in object recognition, thereby improving accuracy of the object detection.