CNN Module Image Segmentation for Embedded Face Analysis
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing CNN-based systems face challenges in performing age and gender estimation on high-resolution face images due to input size constraints of small-scale hardware CNN modules, leading to degraded accuracy and compatibility issues with embedded systems like the Hi3519 SoC.
Innovation Solution
The proposed system divides high-resolution face images into properly sized subimages with judiciously designed overlaps, processes these subimages using a small-scale CNN module, and merges the outputs to generate age and gender classifications, effectively overcoming input size limitations and maintaining accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If high-resolution face images are processed directly by small-scale hardware CNN modules, then image quality is preserved, but the input size exceeds module constraints causing compatibility issues
Solution Approach 1:
The patent divides high-resolution face images into multiple overlapping subimages that can be processed by small-scale CNN modules. Each subimage is processed independently and then merged to produce the final age and gender estimation, enabling compatibility with embedded systems while maintaining accuracy through the overlap region integration
2Adaptability or versatility
If high-resolution face images are downsampled to fit small-scale CNN modules, then compatibility with embedded systems is achieved, but estimation accuracy degrades
Solution Approach 1:
Instead of downsampling the entire high-resolution image, the patent segments it into subimages that fit the module constraints. The overlapping regions between subimages preserve fine details and enable accurate estimation without losing information through aggressive downsampling
Solution Approach 2:
The patent applies different processing strategies to different regions: overlapping regions between subimages are given higher weight during merging to preserve local details and features, while non-overlapping regions use standard processing, thereby maintaining local quality and overall accuracy
3Measurement precision
If multiple overlapping subimages are processed and merged, then accuracy on high-resolution images is maintained, but computational complexity increases
Solution Approach 1:
The patent segments the image into subimages with controlled overlap, allowing parallel processing of multiple smaller units. The merging process combines results from these segments with weighted averaging, achieving high-resolution accuracy while managing complexity through modular processing
Solution Approach 2:
The patent uses a moderate overlap ratio (e.g., 25% overlap) between subimages - enough to preserve boundary details and maintain accuracy, but not so much as to create excessive redundancy and computational burden. This balanced approach optimizes the trade-off between accuracy and complexity
Data Source
AI summary
Embodiments described herein provide various examples of an age and gender estimation system capable of performing age and gender classifications on face images having sizes greater than the maximum number of input pixels supported by a given small-scale hardware convolutional neural network (CNN) module. In some embodiments, the proposed age and gender estimation system can first divide a high-resolution input face image into a set of image patches with judiciously designed overlaps among neighbouring patches. Each of the image patches can then be processed with a small-scale CNN module, such as the built-in CNN module in Hi3519 SoC. The outputs corresponding to the set of image patches can be subsequently merged to obtain the output corresponding to the input face image, and the merged output can be further processed by subsequent layers in the age and gender estimation system to generate age and gender classifications for the input face image.