Inhibition in V1 1. Non-classical receptive field 2. Push pull inhibition

Similar documents
GOLAY COMPLEMENTARY CODES, DOUBLE PULSE REPETITION FREQUENCY TRANSMISSION I. TROTS, A. NOWICKI, M. LEWANDOWSKI J. LITNIEWSKI, W.

Application of Golay Coded Pulse Compression in Air-coupled Ultrasonic Testing of Flexible Package Seal Defect

Thermal Video Analysis for Fire Detection Using Shape Regularity and Intensity Saturation Features

Nobuhiko Wagatsuma, Dr.

THE CONTROL METHOD OF THE INFLOW TURBU- LENCE INTERACTION NOISE FOR ROUTER COOLING FAN

An improved Algorithm of Generating Network Intrusion Detector Li Ma 1, a, Yan Chen 1, b

Research on Decision Tree Application in Data of Fire Alarm Receipt and Disposal

Optimal Control of Induction Heating Processes

Fire Detection System using Matlab

02/11/2015

Impact of Supervisory Gas Pressure on Dry Pipe Sprinkler System Water Delivery Time. Steve Wolin Director, Development and Compliance

Fire Detection Using Image Processing

EE04 804(B) Soft Computing Ver. 1.2 Class 1. Introduction February 21st,2012. Sasidharan Sreedharan

CHAPTER 7 APPLICATION OF 128 POINT FFT PROCESSOR FOR MIMO OFDM SYSTEMS. Table of Contents

Simulation study of evacuation in high-rise buildings

An FT-NIR Primer. NR800-A-006

Ordered Fuzzy ARTMAP: A Fuzzy ARTMAP algorithm with a fixed order

Extreme High-Intensity Sensors Thru-Beam and Fiberoptic Penetrates Many Packaging Materials for Content Verification

Intrusion Detection System based on Speckle Pattern change in Fiber Optic Sensors...Abdulkareem H. Dagher, Shehab A kadhim, Hawraa H.

ELECTRICAL IMPEDANCE TOMOGRAPHY

Automobile Security System Based on Face Recognition Structure Using GSM Network

Advanced Digital Signal Processing Part 4: DFT and FFT

CRUSH: Cognitive Radio Universal Software Hardware

Teaching Landscape Spatial Design with Grading Studies: An Experiment Based on High Fidelity DTM

Low-Frequency Raman Spectroscopy Enabling Affordable Access to the Terahertz Regime

02/11/2015

1. Introduction. Abstract. Keywords: Liquid limit, plastic limit, fall cone, undrained shear strength, water content.

Detector Configurations for the MAJORANA Demonstrator

CURRICULUM VITAE. MARCIA A. FINKELSTEIN Phone: (813) Psychology Department Fax: (813)

IMAGE PROCESSING BASED FIRE DETECTION ANDALERT SYSTEM

2 Specialty Application Photoelectric Sensors

GAMMA OTDR application consists of main window and menu. Using menu user can operate in different modes of application.

Video Smoke Detection using Deep Domain Adaptation Enhanced with Synthetic Smoke Images

Split Range Control for Mobile Room Heating System

Intrusion Detection System: Facts, Challenges and Futures. By Gina Tjhai 13 th March 2007 Network Research Group

Investigation of moisture content fluctuation in mixedflow

Research on the Monitor and Control System of Granary Temperature and Humidity Based on ARM

Fire Dynamics Simulation and Evacuation for a Large Shopping Center (Mall), Part II, Evacuation Scenarios

Experimental Implementation of Spectrum Sensing Using Cognitive Radio

INFLUENCE OF DIFFERENT TEMPERING PERIOD AND VACUUM CONDITIONS ON THE RICE GRAIN BREAKAGE IN A THIN LAYER DRYER

A Study on Landscape Design Paradigm from the Perspective of Visual Impact and Experience

Heat Transfer Enhancement using Herringbone wavy & Smooth Wavy fin Heat Exchanger for Hydraulic Oil Cooling

1066. A self-adaptive alarm method for tool condition monitoring based on Parzen window estimation

On-chip optical detectors in Standard CMOS SOI

Detection system for optic characteristics of automobile glasses

Effects of hearing protection on detection and reaction thresholds for reverse alarms

Fire Detection using Computer Vision Models in Surveillance Videos

Building and Characterizing 14GHz InGaAs Fiber Coupled Photodiodes

Lecture 8 OPENINGS &ORGANIZATION OF FORM AND SPACE

Study of the Relationship between Building Arrangement and Visibility of Open Spaces Based on a Simplified Area Evaluation

Separation of the Pyro- and Piezoelectric Response of Electroactive Polymers for Sensor Applications

Advantages and Disadvantages of Fire Modelling

Digital Urban Simulation

Smoke and Fire Detection

2 General Application Photoelectric Sensors

Performance Comparison of Ejector Expansion Refrigeration Cycle with Throttled Expansion Cycle Using R-170 as Refrigerant

Smoldering Propagation Characteristics of Flexible Polyurethane Foam under Different Air Flow Rates

Construction of a Calibration System for Power Responsivity of Optical detectors Using Different Laser Sources

Proxemic Interactions in Ubiquitous Computing Ecologies

Construction of Wireless Fire Alarm System Based on ZigBee Technology

THE NEXT GENERATION IN VISIBILITY SENSORS OUTPERFORM BOTH TRADITIONAL TRANSMISSOMETERS AND FORWARD SCATTER SENSORS

Prediction of Soil Infiltration Rate Based on Sand Content of Soil

Community & Privacy. Much care should be taken when attempting to establish these intermediate

FiberSystem. Fiber Optic Linear Heat Detection for Special Hazard Applications

Review Article Solar Drying Technology: Potentials and Developments

OVERVIEW AND FUTURE DEVELOPMENT OF THE NEUTRON SENSOR SIGNAL SELF-VALIDATION (NSV) PROJECT

FROSTING BEHAVIOR OF FIN-TUBE HEAT EXCHANGER ACCORDING TO REFRIGERANT FLOW TYPE

Design of CO Concentration Detector Based on Infrared Absorption Method

Operation of a Two-Phase Reverse Loop Thermosyphon

NUMERICAL SIMULATION OF VAPOUR COMPRESSION REFRIGERATION SYSTEM USING REFRIGERANT R152A, R404A AND R600A

Effect of Interaction of Object Color and Lighting Color on a Person s Impression of Interiors

Temperature Control of Heat Exchanger Using Fuzzy Logic Controller

Automatic Lyrics Alignment for Cantonese Popular Music

Range Dependent Turbulence Characterization by Co-operating Coherent Doppler Lidar with Direct Detection Lidar

The Compact Muon Solenoid Experiment. Conference Report. Mailing address: CMS CERN, CH-1211 GENEVA 23, Switzerland

Towards Evolutionary Design Approach: Izola Project

OPTICAL TIME DOMAIN REFELECTOMETER (OTDR): PRINCIPLES

CCTV, HOW DOES IT WORK?

DETECTION AND MONITORING OF ACTIVE FIRES USING REMOTE SENSING TECHNIQUES

Micromegas detectors for the upgrade of the ATLAS Muon Spectrometer

Interferometric optical time-domain reflectometry for distributed optical-fiber sensing

IFIN-HH, Hadron Physics Department infrastructure for ALICE TPC upgrade

Freiburger Materialforschungszentrum SG Material Characterization & Detector Technology. Albert-Ludwigs-Universität Freiburg

LECTURE 11. Dr. Teresa D. Golden University of North Texas Department of Chemistry

ON-LINE SENSOR CALIBRATION MONITORING AND FAULT DETECTION FOR CHEMICAL PROCESSES

Empowering Your Vision

Specific Energy Consumption comparative study of Hot Air dryer and Heat Pump dryer for highland drying process Sayompon Srina

Available online Journal of Scientific and Engineering Research, 2018, 5(11): Research Article

EFFECT OF COMPACTION ON THE UNSATURATED SHEAR STRENGTH OF A COMPACTED TILL

Band Selection & Algorithm Development for Remote Sensing of Wildland Fires

2 General Application Photoelectric Sensors

Improving Heating Performance of a MPS Heat Pump System With Consideration of Compressor Heating Effects in Heat Exchanger Design

GOES-R ABI Emissive IR Bands Radiometric Performance Monitoring and Trending

When Precision is not good enough

Preliminary Exploration: Fault Diagnosis of the Circulating-water Heat Exchangers based on Sound Sensor and Non-destructive Testing Technique

THE LANDSCAPE ECOLOGICAL ASSESSMENT MODEL AND ITS APPLICATIONS

A Method for Predicting the Matric Suction of Unsaturated Soils with Soil Color Recognition

Benefits of Enhanced Event Analysis in. Mark Miller

Experimental Study on Utilization of E -Waste in Cement Concrete

Recent BRANZFIRE enhancements and validation

Transcription:

BICV Summer School, Shenyang, China, 2015 1 Inhibition in V1 1. Non-classical receptive field 2. Push pull inhibition George Azzopardi Nicolai Petkov

BICV Summer School, Shenyang, China, 2015 2 1.Non-classical receptive field inhibition

BICV Summer School, Shenyang, China, 2015 3

BICV Summer School, Shenyang, China, 2015 4

BICV Summer School, Shenyang, China, 2015 5

BICV Summer School, Shenyang, China, 2015 6

BICV Summer School, Shenyang, China, 2015 7

BICV Summer School, Shenyang, China, 2015 8

BICV Summer School, Shenyang, China, 2015 9

BICV Summer School, Shenyang, China, 2015 10 Visual perception [Solomon & Pelli, 1994]

BICV Summer School, Shenyang, China, 2015 Visual perception [Petkov & Westenberg, 2003] 11

BICV Summer School, Shenyang, China, 2015 12

BICV Summer School, Shenyang, China, 2015 13

BICV Summer School, Shenyang, China, 2015 14

BICV Summer School, Shenyang, China, 2015 15 Visual perception [Galli & Zama, 1931]

BICV Summer School, Shenyang, China, 2015 16 Visual perception Orientation-contrast pop-out

BICV Summer School, Shenyang, China, 2015 17 Visual perception

BICV Summer School, Shenyang, China, 2015 18 Neurophysiology Receptive field and response of an orientation [Hubel & Wiesel, 1962] selective neuron

BICV Summer School, Shenyang, China, 2015 19 Neurophysiology Impulse response of an orientation selective neuron modelled by a 2D Gabor function [Daugman, 1985]

BICV Summer School, Shenyang, China, 2015 20 What is the neurophysiology of this visual perception?

BICV Summer School, Shenyang, China, 2015 21 Neurophysiology CRF Inhibition surround

BICV Summer School, Shenyang, China, 2015 22 Neurophysiology [Blakemore and Tobin, 1972]

BICV Summer School, Shenyang, China, 2015 23 Link to perception

BICV Summer School, Shenyang, China, 2015 24 Neurophysiology Mean firing rate [spikes/s] 60 40 20 Orientation-contrast cell [Nothdurft et. al, 1999]

BICV Summer School, Shenyang, China, 2015 25 Link to perception

BICV Summer School, Shenyang, China, 2015 26 Neurophysiology Mean firing rate [spikes/s] 60 40 20 (a) (b) (c) (d) General suppression cell [Nothdurft et. al, 1999]

BICV Summer School, Shenyang, China, 2015 27 Link to perception

BICV Summer School, Shenyang, China, 2015 28 Computational models

BICV Summer School, Shenyang, China, 2015 29 Gabor filter

BICV Summer School, Shenyang, China, 2015 30 Gabor filter Kernel: 27

BICV Summer School, Shenyang, China, 2015 31 Gabor filter Kernel: Filter:

BICV Summer School, Shenyang, China, 2015 32 Gabor energy filter

BICV Summer School, Shenyang, China, 2015 33 Gabor energy filter Gabor energy:

BICV Summer School, Shenyang, China, 2015 34 Gabor energy filter input I output E 0 o output E 90 o max(e 0 o,e 90 o )

BICV Summer School, Shenyang, China, 2015 35 Surround suppression CRF Inhibition surround

BICV Summer School, Shenyang, China, 2015 36 Surround suppression Different effects of the surround

BICV Summer School, Shenyang, China, 2015 37 Surround weighting function

BICV Summer School, Shenyang, China, 2015 38 Surround weighting function

BICV Summer School, Shenyang, China, 2015 39 Surround weighting function

BICV Summer School, Shenyang, China, 2015 40 Orientation-contrast suppression Mean firing rate [spikes/s] 60 40 20

BICV Summer School, Shenyang, China, 2015 41 Orientation-contrast suppression term

BICV Summer School, Shenyang, China, 2015 42 Orientation-contrast suppression term Input I Gabor energy E 0 o Suppression S 0 o term

BICV Summer School, Shenyang, China, 2015 43 Orientation-contrast suppression filter

BICV Summer School, Shenyang, China, 2015 44 Orientation-contrast suppression filter

BICV Summer School, Shenyang, China, 2015 45 Orientation-contrast suppression

BICV Summer School, Shenyang, China, 2015 46 Back to visual perception

BICV Summer School, Shenyang, China, 2015 47 Back to visual perception 38

BICV Summer School, Shenyang, China, 2015 48 Computer vision application

BICV Summer School, Shenyang, China, 2015 49 Texture suppression Input Gabor energy With suppression

BICV Summer School, Shenyang, China, 2015 50 Contour enhancement by texture suppression Input Gabor energy With suppression

BICV Summer School, Shenyang, China, 2015 51 Binary contour maps Input Edge strength Binarized response

BICV Summer School, Shenyang, China, 2015 52 Desired output Input image and associated desired output Dataset: www.cs.rug.nl/ imaging

BICV Summer School, Shenyang, China, 2015 53 Performance measure Desired output Operator output

BICV Summer School, Shenyang, China, 2015 54 Performance measure

BICV Summer School, Shenyang, China, 2015 55 Canny performance Input Desired output Best Canny P = 0.23

BICV Summer School, Shenyang, China, 2015 56 Suppressed Gabor performance Input Desired output Suppressed Gabor P = 0.42

BICV Summer School, Shenyang, China, 2015 57 Performance comparison Input Best Canny Suppressed Gabor P = 0.23 P = 0.42

BICV Summer School, Shenyang, China, 2015 58 Performance comparison Input Best Canny Suppressed Gabor P = 0.14 P = 0.34 [Grigorescu et al., 2003]

BICV Summer School, Shenyang, China, 2015 59 Performance comparison 1 0.8 Goat 3 Elephant 2 Hyena Gazelle 2 0.6 Performance 0.4 0.2 0 C G B1 B2 C G B1 B2 C G B1 B2 C G B1 B2 C G B1 B2 (C) Canny, (G) Gabor, (B1) orientation contrast suppression, (B2) general suppression

BICV Summer School, Shenyang, China, 2015 60 Beyond receptive fields and Gabor functions

BICV Summer School, Shenyang, China, 2015 61 Beyond receptive fields and Gabor functions Canny edge detector

BICV Summer School, Shenyang, China, 2015 62 Beyond receptive fields and Gabor functions Canny edge detector with surround suppression

BICV Summer School, Shenyang, China, 2015 63 Canny with surround suppression Input Canny With suppression [Grigorescu et al., 2004]

BICV Summer School, Shenyang, China, 2015 64 Split surround

BICV Summer School, Shenyang, China, 2015 65 Input Best Canny Suppressed Gabor

BICV Summer School, Shenyang, China, 2015 66 Multi-scale Canny with split surround suppression Input Suppressed Canny (multi-scale) Desired output [Papari et al., 2006]

BICV Summer School, Shenyang, China, 2015 67 2. Push-pull inhibition

BICV Summer School, Shenyang, China, 2015 68 Simple Cells receive push-pull inhibition (Anderson et al., 2000) Stimuli of the reverse contrast evoke responses of the opposite sign With push-pull inhibition linear models achieve contrast invariant orientation tuning CORF model (without inhibition) Already achieves contrast invariant orientation tuning Only responds to preferred orientation and contrast We explore further properties

BICV Summer School, Shenyang, China, 2015 69 Push-Pull inhibition RFs arrangement of LGN cells with preferred polarity RFs arrangement of LGN cells with opposite polarity push (excitation) pull (inhibition) Preferred stimulus

BICV Summer School, Shenyang, China, 2015 70 Overlap index (Martinez et al., 2005)

BICV Summer School, Shenyang, China, 2015 71 Overlap Index 1 β = 0 Normalized response to preferred orientation 0.57 0.19 0.04 β = 2 β = 4 β = 6 90 100 106 120 Orientation bandwidth (degrees)

BICV Summer School, Shenyang, China, 2015 72 Response of CORF with push pull inhibition

BICV Summer School, Shenyang, China, 2015 73 Response of CORF with push pull inhibition r S (x, y )

BICV Summer School, Shenyang, China, 2015 74 Response of CORF with push pull inhibition r S (x, y ) r Sβ (x, y ) + β

BICV Summer School, Shenyang, China, 2015 75 Response of CORF with push pull inhibition r S (x, y ) kr Sβ (x, y ) + β

BICV Summer School, Shenyang, China, 2015 76 Signal-to-noise Ratio Edge stimulus Band-limited texture

BICV Summer School, Shenyang, China, 2015 77 Signal-to-noise Ratio + = Edge stimulus Band-limited texture Test image

BICV Summer School, Shenyang, China, 2015 78 Signal-to-noise ratio Input Output of models CORF CORF+PP SNR= 20.35 SNR= 31.54 SNR= 9.45 SNR= 23.83

BICV Summer School, Shenyang, China, 2015 79 Signal-to-noise ratio Input Output of models CORF CORF+PP SNR= 12.06 SNR= 28.68 SNR= 2.83 SNR= 14.73

BICV Summer School, Shenyang, China, 2015 80 Signal-to-noise ratio Input Output of models CORF CORF+PP SNR= 17.57 SNR= 21.52 SNR= 5.85 SNR= 8.18

BICV Summer School, Shenyang, China, 2015 81 Separability of spatial frequency and orientation Real simple cell (Webster, De Valois, et al., 1985). CORF without inhibition Fundamental amplitude 50 40 30 20 10 Orientation ( ) 5 20 35 50 65 0 0.17 0.33 0.67 1.33 2.67 5.34 Spatial Frequency (cycles/degree) Normalized response 1 Orientation ( ) 0 10 25 30 35 0 2 4 8 16 32 Spatial Frequency

BICV Summer School, Shenyang, China, 2015 82 Separability of spatial frequency and orientation Real simple cell (Webster, De Valois, et al., 1985) CORF with push-pull inhibition Fundamental amplitude 50 40 30 20 10 Orientation ( ) 190 200 215 220 225 0 0.17 0.33 0.67 1.33 2.67 5.34 Spatial Frequency (cycles/degree) Normalized response 1 Orientation ( ) 0 10 25 30 35 0 2 4 8 16 32 Spatial Frequency

BICV Summer School, Shenyang, China, 2015 83 Spatial frequency tuning sensitive to contrast (Sceniak et al., 2002) Normalized response 1 0.5 0 CORF without inhibition 1 2 4 8 16 32 Spatial Frequency CORF with push-pull inhibition Contrast 80% 20% 1 2 4 8 16 32 Spatial Frequency

BICV Summer School, Shenyang, China, 2015 84 Contour detection Input image CORF with push-pull inhibition CORF without inhibition

BICV Summer School, Shenyang, China, 2015 85 Contour detection Input image CORF with push-pull inhibition Gabor with isotropic surround suppression

BICV Summer School, Shenyang, China, 2015 86 Contour detection Input image CORF with push-pull inhibition Canny

BICV Summer School, Shenyang, China, 2015 87 Contour detection Input image CORF with push-pull inhibition CORF without inhibition

BICV Summer School, Shenyang, China, 2015 88 Contour detection Input image CORF with push-pull inhibition Gabor with isotropic surround suppression

BICV Summer School, Shenyang, China, 2015 89 Contour detection Input image CORF with push-pull inhibition Canny

BICV Summer School, Shenyang, China, 2015 90 Results: CORF with push-pull inhibition vs others Right-tailed paired t -test RuG: 40 images CORF without inhibition: p < 0.01 Gabor with surround suppression: p < 0.01 Canny: p < 0.01 Berkeley: 500 images CORF without inhibition: p < 0.01 Gabor with surround suppression: p < 0.01 Canny: p < 0.01

BICV Summer School, Shenyang, China, 2015 91 Conclusions Computational modelling of two types of inhibition Gabor à non-classical receptive field CORF à push-pull inhibition CORF+PP improves SNR CORF+PP is more effective than Gabor with surround suppression in contour detection

BICV Summer School, Shenyang, China, 2015 Outlook Combine non-classical receptive field and push-pull inhibition in one model? Contour detection is expected to be improved even further Incorporate this model for object segmentation in real images The state-of-the-art for segmentation algorithms use orientation- selective filters without inhibition 92

BICV Summer School, Shenyang, China, 2015 93 References G. Azzopardi and N. Petkov, A CORF computational model that relies on LGN input outperforms the Gabor function model, Biological Cybernetics, 106(3), 177-189, 2012. G. Azzopardi, A. Rodriguez-Sanchez, J. Piater, N. Petkov, A computational CORF model of a simple cell with push-pull inhibition exhibits improved SNR and outperforms state-of-the-art contour detection operators, to be submitted.

BICV Summer School, Shenyang, China, 2015 94 Matlab Code http://matlabserver.cs.rug.nl/

BICV Summer School, Shenyang, China, 2015 95 Matlab Code http://matlabserver.cs.rug.nl/ Thank you! 谢谢 Grazzi! Dank jullie well!

Webster, Michael A, Russell L De Valois, et al. (1985). Relationship between spatial-frequency and orientation tuning of striate-cortex cells.in: JOSA A 2.7, pp. 1124 1132. References BICV Summer School, Shenyang, China, 2015 96 Anderson, JS et al. (2000). Orientation tuning of input conductance, excitation, and inhibition in cat primary visual cortex.in: Journal of Neurophysiology 84.2, 909 926. Blakemore, C and EA Tobin (1972). Lateral inhibition between orientation detectors in cats visual-cortex.in: Experimental Brain Research 15.4, 439 &. ISSN: 0014-4819. Daugman, J. G. (1985). Uncertainty Relation for Resolution in Space, Spatial-Frequency, and Orientation Optimized by Two-Dimensional Visual Cortical Filters.In: Journal of the Optical Society of America a-optics Image Science and Vision 2.7, pp. 1160 1169. Hubel, D. H. and T. N. Wiesel (1962). Receptive Fields, binocular interaction and functional architecture in cats visual cortex.english. In: J.Physiol.(Lond.) 160.1, pp. 106 154. Jones, J. P. and L. A. Palmer (1987). An Evaluation of the Two-Dimensional Gabor Filter Model of Simple Receptive-Fields in Cat Striate Cortex.In: Journal of Neurophysiology 58.6, pp. 1233 1258. Marr, D. and E. Hildreth (1980). Theory of Edge-Detection.In: B-Biological Sciences 207.1167, pp. 187 217. Proceedings of the Royal Society of London Series Martinez, LM et al. (2005). Receptive field structure varies with layer in the primary visual cortex.in: Neuroscience 8.3, 372 379. Nature Nothdurft, HC et al. (1999). Response modulation by texture surround in primate area V1: Correlates of popout under anesthesia.in: Visual Neuroscience 16.1, 15 34. Petkov, N and MA Westenberg (2003). Suppression of contour perception by band-limited noise and its relation to nonclassical receptive field inhibition.in: Biological Cybernetics 88.3, 236 246. Sceniak, Michael P et al. (2002). Contrast-dependent changes in spatial frequency tuning of macaque V1 neurons: effects of a changing receptive field size.in: Journal of Neurophysiology 88.3, pp. 1363 1373.

BICV Summer School, Shenyang, China, 2015 97 Hysteresis Thresholding