Bio-inspired Object Recognition via Human Visual Pathway Emulation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Geometric facial recognition algorithms require high-resolution images, are sensitive to noise, and struggle with partial or distorted images, necessitating large sample sets for accurate identification, which limits their effectiveness in dynamic environments.
Innovation Solution
A bio-inspired model emulating the human visual pathway, using saccadic eye movements and components like the retina, fovea, and lateral geniculate nucleus to extract features from images, allowing for efficient recognition even in low-resolution or partially occluded images by generating blocks and applying PCA and feature extraction algorithms.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If geometric facial recognition algorithms are used to extract landmarks and analyze relative position, size, and shape of facial features, then object recognition accuracy is improved, but the system requires high-resolution images and is sensitive to noise
Solution Approach 1:
The patent segments the facial recognition process into multiple stages: initial landmark detection, feature extraction, and verification. By dividing the recognition process into discrete steps with intermediate validation, the system can identify and correct errors caused by noise without requiring perfect high-resolution input throughout the entire process.
Solution Approach 2:
The system dynamically adjusts recognition parameters such as landmark tolerance thresholds, feature weighting, and matching criteria based on image quality assessment. When noise is detected or resolution is insufficient, the algorithm modifies these parameters to maintain recognition accuracy without being overly sensitive to degraded input conditions.
2Measurement precision
If geometric algorithms require complete pictures of subjects to determine relative position, size, and shape of facial features, then measurement precision is improved, but the system cannot handle partial or occluded images
Solution Approach 1:
The patent implements partial action by enabling facial recognition to function with incomplete facial data. The system can perform recognition using only visible portions of the face, extracting sufficient features from available regions rather than requiring the complete face. This allows operation with partially occluded images while maintaining acceptable accuracy through adaptive feature selection and weighting.
3Measurement precision
If geometric facial recognition algorithms use large sets of sample images for comparison, then object recognition accuracy is improved, but device complexity and processing time increase
Solution Approach 1:
The system extracts and utilizes only the most discriminative and informative features from sample images for comparison, rather than processing entire images or all possible features. By selecting and weighting key facial landmarks and features that provide the most identification value, the system achieves accurate recognition with reduced computational requirements and smaller effective sample representations.
4Adaptability or versatility
If geometric algorithms translate subject images to fit sample image dimensions, then compatibility is improved, but recognition accuracy deteriorates due to distortion
Solution Approach 1:
The patent employs dynamic scaling and transformation that adapts to each subject image's characteristics rather than applying fixed translation rules. The system dynamically adjusts transformation parameters based on detected facial landmarks and image geometry, preserving relative feature relationships and proportions while achieving compatibility with standard processing dimensions, thereby maintaining accuracy across varied input formats.
Data Source
AI summary
Methods, systems, and storage media are described for detecting objects in image data are provided. In embodiments, a computing device may generate a plurality of blocks from a captured image. Each block may represent a corresponding region of the captured image. The computing device may extract a first feature from at least one block of the plurality of blocks; determine a second feature based at least on the first feature; and determine a third feature based on the second feature. The computing device may determine, as a matching image, a stored image from among a plurality of stored images including an object that has greatest maximum correlation with the third feature. Other embodiments may be described and/or claimed.