Image Recognition Device Using Projection Vector Folding
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current image recognition devices face challenges in reducing arithmetic operations and memory requirements due to the large number and order of support vectors needed for image classification, making real-time processing difficult and inefficient.
Innovation Solution
The image recognition device employs a method that reduces the order of support vectors through a folding operation, generating a projection vector and converting training vectors into two-dimensional vectors, which are then used to generate training parameters for classification, thereby reducing the number of multiplying operations and memory needed.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a large number of support vectors and high-order support vectors are used to improve image classification performance, then recognition accuracy is improved, but the number of arithmetic operations and memory requirements increase significantly
Solution Approach 1:
The patent extracts only the essential features from high-order support vectors by projecting them onto a lower-dimensional space using a projection vector. This extraction process removes redundant information while preserving the discriminative features needed for accurate classification, thereby reducing computational complexity without significantly compromising recognition accuracy
Solution Approach 2:
The patent transforms high-order support vectors from a high-dimensional space into lower-dimensional vectors through a projection operation. This dimensionality reduction maps the original support vectors onto a new space with fewer dimensions, reducing the computational burden of arithmetic operations and memory requirements while maintaining the essential classification information
2Reliability
If the order N of each support vector is increased to several thousands to improve classification capability, then recognition performance is improved, but the multiplying operations needed reach N×N levels requiring millions of operations
Solution Approach 1:
The patent extracts the essential classification information from high-order support vectors by projecting them onto a lower-dimensional space. This extraction reduces the vector order from thousands to a manageable size, eliminating the need for millions of multiplying operations while preserving the core classification capability
Solution Approach 2:
The patent changes the parameter of support vector order from high (thousands) to low (reduced dimension) through a projection operation. This parameter transformation maintains the essential classification information while dramatically reducing the computational complexity from O(N×N) to a much lower order, enabling real-time processing
3Measurement precision
If a large number of support vectors (several hundreds through several thousands) are stored to improve classification accuracy, then recognition performance is improved, but memory requirements and hardware complexity increase making real-time processing difficult
Solution Approach 1:
The patent extracts essential features from a large number of high-order support vectors by projecting them onto a lower-dimensional space. This extraction reduces both the number of vectors needed and their dimensionality, significantly decreasing memory requirements and enabling real-time processing while maintaining classification accuracy
Solution Approach 2:
The patent transforms the storage and processing of support vectors from high-dimensional space to lower-dimensional space through projection. This dimensionality change reduces the memory footprint from storing thousands of high-order vectors to storing fewer low-order vectors, making real-time processing feasible
Data Source
AI summary
An image recognition device in accordance with the inventive concept may include an input vector extraction part extracting an input vector from an input image; a compression vector conversion part converting the input vector into a compression vector using a projection vector; a training parameter generation part receiving a training vector to generate a training parameter using a projection vector obtained through a folding operation of the training vector; and an image classification part classifying the compression vector using the training vector to output image recognition data.