Computing with Biologically Inspired Neural Oscillators: Application to Color Image Segmentation

Transcription

1 Computing with Biologically Inspired Neural Oscillators: Application to Color Image Segmentation Authors: Ammar Belatreche, Liam Maguire, Martin McGinnity, Liam McDaid and Arfan Ghani Published: Advances in Artificial Intelligence, Volume 2010, Article ID Presented By: Kyle Stever

2 Objective Attempt to mimic human brain for image processing Partitioning of an image into homogenous regions Applications in: Pattern recognition Image processing Robotics Attempted using biologically inspired neural oscillators

3 Biologically Inspired Oscillators: History Observed in a variety of biological phenomena Heartbeat Neural activity Visual cortices Discovered in 1926 by Van Der Pol While studying the characteristics of a triode circuit that showed self-sustained oscillations Used in 1928 by Van Der Pol and Van Der Mark To describe the heartbeat and present an electrical model Mathematically described in 1963 by Hodgkin and Huxley To describe neuron membrane electric transmission

4 Biologically Inspired Oscillators: Characterizations Slow capacitor charge and quick discharge Period of oscillation proportional to the time needed to charge the capacitor Relaxation time Oscillations are stimulus-dependent Synchronization of oscillations emerge when stimuli appear to belong to coherent object No synchrony of oscillations when stimuli appears to belong to differing objects

5 Single Relaxation Oscillator Dynamics: Nullclines A single neural oscillator X and y-nullclines represented in the phase plane

6 Single Relaxation Oscillator Dynamics: Temporal Activity Positive input: a, b, d, e) Temporal activity of units x and y c, f) Convergence to limit cycle in phase plane Negative input: a) Temporal activity of units x and y b) Convergence to a fixed point in phase plane

7 Coupled Relaxation Oscillator Dynamics: Nullclines Two coupled neural oscillators Nullclines of two coupled oscillators and the convergence of their temporal activity to a synchronized limit cycle

8 Coupled Relaxation Oscillator Dynamics: Temporal Activity Synchronization in a pair of oscillators due to strong coupling strength Desynchronization in a pair of oscillators due to weak coupling strength

9 Grey Scale Image Segmentation: Network Architecture A grid of competitive self-organizing locally excitatory globally inhibitory neural oscillators between each neural oscillator Connections between each neural oscillator

10 Grey Scale Image Segmentation: Simulation small image 10x10 pixel grey scale image Temporal activity of neural oscillators Temporal activity of neural oscillators Neuron indices firing at various times

11 Grey Scale Image Segmentation: Computational Time Reduction X and y-nullclines represented in the phase plane

12 Grey Scale Image Segmentation: Simulation large image 110x90 pixel grey scale image segmented into different regions

13 Color Image Segmentation: Challenge Challenge between color and black and white Greyscale image the pixel values could be directly used as features Color image the pixel values are encoded by a triplet of integer values Cannot be directly represented by a single oscillator Solution Feed the feature sets through a dimensionality reduction process

14 Color Image Segmentation: Kohonen Self-Organising Maps Select a random input vector Traverse each node in the map Use Euclidean distance formula to find similarity between the input vector and the map's node's weight vector Track the node that produces the smallest distance Update the nodes in the neighbourhood of BMU by pulling them closer to the input vector Wv(t + 1) = Wv(t) + Θ(t)α(t)(D(t) - Wv(t)) Node Update Procedure Increment t and repeat from 2 while t < λ Example of SOM on 3 and 8 color image

15 Color Image Segmentation: HSV Color Space Hue Represents color Saturation Range of grey Value Range of brightness HSV Color Space Diagram

16 Color Image Segmentation: Spatial Features Extraction of local spatial features.

17 Color Image Segmentation: Feature Vector Feature vector for each pixel, using both spatial and chromatic characteristics

18 Color Image Segmentation: Kohonen SOM Application Kohonen s self-organizing map

19 Color Image Segmentation: System Overview Neuro-inspired framework for color image segmentation

20 Color Image Segmentation: Simulation 130x80 pixel image segmented into different color regions

21 Color Image Segmentation: Simulation 100x129 pixel image segmented into different color regions

22 Color Image Segmentation: Simulation 128x128 pixel image segmented into different color regions

23 Evaluation Analysis Subjectively provided good segmentation Currently no defined objective value of segmentation quality Advantages: Based on human neurology Parallel computing Robustness

24 References Computing with Biologically Inspired Neural Oscillators: Application to Color Image Segmentation Advances in Artificial Intelligence, Volume 2010, Article ID Ammar Belatreche, Liam Maguire, Martin McGinnity, Liam McDaid and Arfan Ghani Accessed last on February 12, d/som.html Accessed last on February 12, Accessed last on February 12, 2011