Edge Detection & Form/Motion Interaction

Edge Detection & Form/Motion Interaction VISN2211 Sieu Khuu David Lewis

Part 1: Edge detection • Finding the lines within an image that indicate structure or form. • This is done by finding those points which there are abrupt luminance changes. • Being able to identify the forms of objects in the environment.

Convolution • A mathematical operation on two functions which will produce a third function. • The third function is a modified version of the first. • In edge detection we use convolution to extract the edges from an image.

First Order Edge Operators • Past research have applied what are known as first order edge operators. • These work by comparing the relative luminance intensities within the profiles of small areas to determine if there is an edge at that location. Horizontal edge operator Diagonal edge operator +1 +1 Vertical edge operator -1 -1 +1 -1

Luminance Profiles • In a luminance profile a steep slope indicates an edge. • These can be broken down into two derivative functions. • Current edge detection methods involve the 2nd derivative of a luminance profile. • Zero-crossing edge +1 Luminance 0 horizontal location +1 edge 1st Derivative non-edge 0 +1 Edge (crosses y=0) 0 2nd Derivative -1

Natural Images

Horizontal Edges +1 -1

Vertical Edges -1 +1

Diagonal Edges +1 -1

-1 -1 -1 -1 +8 -1 -1 -1 -1 -1 -1 -1 +1 +1 +1 -1 +1 +1 +1 -1 +1 +1 -1 -1 -1 Second Order Edge Operators • First order operators fail to detect many of the edges when used individually. • To detect all possible edges, multiple first order operators must be used. This is inefficient. • Second order edge operators can be thought of as combinations of multiple first-orders. • The most complex of which is the omnidirectional: + + + + + + + =

-1 -1 -1 -1 +8 -1 -1 -1 -1 Omnidirectional Edges

Noise • As previously shown, zero-crossing edge operators produce a lot of noise. • Noise = false edges • A more efficient (less noisy) method of edge detection is required • More complex edge operators have been produced in research. • Canny • Sobel

Canny edge operator • 5x5 Gaussian filter • 1/159[2, 4, 5, 4, 2; 4, 9, 12, 9, 4; 5, 12, 15, 12, 5; 4, 9, 12, 9, 4; 2, 4, 5, 4, 2] 2 4 5 4 2 4 9 12 9 4 5 12 15 12 5 4 9 12 9 4 2 4 5 4 2

Sobel edge operator • Two 3x3 kernels • [1, 2, 1; 0, 0, 0; -1, -2, -1] horizontal • [1, 0, -1; 2, 0, -2; 1, 0, -1] vertical horizontal +1 +2 +1 vertical 0 0 0 +1 0 -1 -1 -2 -1 +2 0 -2 +1 0 -1

Edge Detection Summary • The visual system needs an efficient way to detect edges in order to determine form. • First order edge operators are inefficient due to noise (false edges). • The visual system must use a more complex algorithm. • I.e. Sobel edge operator.

Part 2: Form/Motion Interaction • The visual system is a collection of functional processes. • These processes analyse and code different properties of an image. • Color • Form • Motion • Etc. • Current research suggests that there are many interactions between these processes. • I.e. motion perception affecting form perception

The Visual System’s Processes Rees, G., Kreiman, G., & Koch (2002)

Form and motion interaction • A well studied case of form and motion interaction is the influence of image motion on the computation of space and spatial position. • For example: • The flash-lag-effect: The perceived position of a briefly flashed object appears to perceptually lags behind a continuously moving object (MacKay, 1958). • McFarland Illusion (McFarland, 1970): Apparent motion (phi motion) influences the space and position of elements in close proximity. • Vernier judgments of position (DeValois & De Valois 1992): The perceived position of a stationary object with internal motion appears shifted in the direction of motion. • These illusions clearly indicate that the derivation of form is heavily influenced by image motion. • Today’s experiment will be similar to DeValois & DeValois, 1992.

Demonstration of form and motion interaction using MAE

Edge Detection & Form/Motion Interaction