Presentation on theme: "Camouflage Detection An introduction Presented by: Ani Starrenburg."— Presentation transcript:
Camouflage Detection An introduction Presented by: Ani Starrenburg
General Camouflaging Strategies Cryptic Camouflage Little Button Quail Traditional US Army Camouflage Pattern
General Camouflaging Strategies Mimicry Dronefly Rose Greenbow, Confederate Spy
General Camouflaging Strategies Disruption Sumatran Tiger Dazzle Camouflage
General Camouflaging Strategies Countershading Impala Non-Countershaded Warship
Translucence/Transparency SeawaspInvisibility Cloak General Camouflaging Strategies
Detecting Camouflaged Objects:
Camouflage Detection Methods Standard Object Detection Methods Edge Detection Models Contrast Energy Detection Model Motion Detection Correlation Models Gradient Models Energy Models
Edge Detectors: Gradient Laplacian With Gaussian Gaussian
Canny Detector Optimal Edge Detector Multiple Stage Algorithm Perform Gaussian smoothing Find edge strengths |G| = |Gx| + |Gy| Detection of edge direction theta = invtan(Gy/Gx) Relate edge direction to a direction that can be traced in an image Apply non-maximum suppression Use hysteresis to eliminate streaking
LaPlacian or LoG Smooth with a Gaussian mask Calculate the second derivatives Search for zero crossings Or Convolve the image with the Laplacian of the Gaussian
Contrast Energy (CE) Model Uses the output signal from similarly-oriented odd o[x] and even e[x] filters. Energy function is defined as: E 2 (x) = e 2 (x) + o 2 (x) Always positive Shows high output when o(x), e(x) or both are high.
Camouflage Detection Methods to be Discussed Convexity-Based Detection – exploits the principle of countershading to detect camouflaged objects Texture Detection – intensive texture analysis distinguishes camouflaged object from background. Also, uses Canny detector to bring up edges
Motion Breaks Camouflage Region of common velocity is perceived As a unit and stands out against the static background
Reichardt Correlation Model Computes motion as the ratio of the partial derivatives of the input image brightness with respect to space and time. Two spatially-separate detectors. Output of one of the detectors is delayed. The two outputs are multiplied to determine if there is a correlation.
Multichannel Gradient Model Uses multiple channels of higher derivatives The more derivatives used lowers the chance of that all will be zero at the same time Uses a least sqaures approximation of the derivatives
Motion Energy Model Uses two sets of oriented detectors(leftwards and rightwards), each composed of an odd and an even filter. Energy is calculated by summing the squares of the two similarly-oriented filters. Calculate opponent energy (difference of leftward and rightward results) Normalize by dividing by static energy to give velocity estimates
An aside: Research on Active Camouflage Animals that can escape edge detection Animals that can camouflage motion
To Do List: Apply edge detectors and contrast energy detectors to camouflaged and illusory images and view results. Research visual models developed from observing animal behavior and development. Research studies in psychology for further understanding of vision process.
Is there a core visual system? CAMOUFLAGECAMOUFLAGE ARTART
Bibliography Motion Illusions and Active Camouflage, Lewis Dartnell,http://www.ucl.ac.uk/~ucbplrd/motion/motion_middle.htmlhttp://www.ucl.ac.uk/~ucbplrd/motion/motion_middle.html Canny Edge Detection Tutorial, Bill Green, Honeybee, Ground-dwelling birds, Sumatran tiger, Biomimicry, Countershading, Translucence, Canny Edge Detection, Optical Camouflage, Multi-Channel Gradient Model,