Spiking Neuron Network Salient Feature Detection via Latency Encoding
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing approaches for detecting salient features in visual inputs, particularly in the presence of variable brightness and color, are inadequate in terms of temporal and spatial response, and spiking neuron networks used for visual attention are often overly complex and slow to adapt to changing conditions.
Innovation Solution
A computerized neuron-based network that encodes visual inputs using spiking neuron networks, where neurons encode attributes into pulse latencies and inhibit responses based on latency windows, allowing for the detection of salient features by suppressing non-salient information through a 'winner-takes-all' methodology.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Speed
If spiking neuron networks are used to encode visual information, then temporal and spatial response is improved, but device complexity increases
Solution Approach 1:
The network is divided into distinct functional layers: feed-forward excitatory neurons for rapid response, lateral excitatory neurons for spatial encoding, and inhibitory neurons for competitive selection. This segmentation allows each component to specialize in specific tasks, improving overall temporal response while managing complexity through functional decomposition
Solution Approach 2:
Inhibitory neurons act as intermediaries that mediate competition between different feature detectors. These neurons receive input from multiple excitatory neurons and provide feedback inhibition, enabling the network to select salient features without requiring complex direct inhibition circuits between all excitatory neurons
2Measurement precision
If spiking neuron networks are used for visual attention, then detection precision is improved, but response time increases
Solution Approach 1:
The feed-forward excitatory neurons are pre-configured with stimulus-responsive properties that enable immediate response to visual input. Their direct connection from input to output layers, without requiring iterative processing, allows preliminary detection of salient features before inhibitory feedback is applied, thus maintaining fast response time while achieving precise detection
Solution Approach 2:
The network employs periodic pulse trains with specific latency patterns to encode different visual attributes. By using rhythmic spiking patterns rather than continuous signaling, the network achieves precise temporal encoding of feature salience while maintaining efficient energy usage and faster overall response compared to continuous processing
3Loss of information
If multiple attributes are encoded into pulse latencies, then information content is improved, but processing complexity increases
Solution Approach 1:
Multiple visual attributes (orientation, spatial frequency, color) are merged into a single temporal dimension through pulse latency encoding. Different attributes are represented by different latency values within the same neural population, eliminating the need for separate neural pathways for each attribute and reducing overall processing complexity while preserving full information content
Solution Approach 2:
The network transforms spatial and spectral information into the temporal dimension by encoding attribute values as pulse latencies. This dimensional transformation allows multiple attributes to be represented simultaneously in a single neural response, converting a multi-dimensional encoding problem into a temporal sequencing problem that is more efficiently processed by the spiking network
4Measurement precision
If inhibition is applied based on latency windows, then signal-to-noise ratio is improved, but processing speed decreases
Solution Approach 1:
The inhibitory mechanism applies partial inhibition only to neurons whose pulse latencies fall outside the salient feature window, rather than uniformly inhibiting all neurons. This selective partial action maintains fast processing for salient features while applying noise suppression only where needed, achieving improved signal-to-noise ratio without uniformly reducing processing speed across the entire network
Data Source
AI summary
Apparatus and methods for detecting salient features. In one implementation, an image processing apparatus utilizes latency coding and a spiking neuron network to encode image brightness into spike latency. The spike latency is compared to a saliency window in order to detect early responding neurons. Salient features of the image are associated with the early responding neurons. A dedicated inhibitory neuron receives salient feature indication and provides inhibitory signal to the remaining neurons within the network. The inhibition signal reduces probability of responses by the remaining neurons thereby facilitating salient feature detection within the image by the network. Salient feature detection can be used for example for image compression, background removal and content distribution.