Fuzzy Inferences for Edge Detection

Fuzzy inferences System Appliedto edge detection as a part of object identification in Video : 

Fuzzy inferences System Appliedto edge detection as a part of object identification in Video Presented By Brijeshkumar Shah Y7104071

Objective : 

Objective Why and What is Edge Detection ? Why and How Fuzzy Inferences System is used ? The FIS used for the project. Results on Image. Results on Video. Limitations of the system.

Why Edge Detection is required ? : 

Object Identification Video compression Gesture Recognition Foreground Extraction Different methods available Edge detection is one of the simplest method Edges defines object and for series of frame motion in object can be found. Why Edge Detection is required ?

What is Edge Detection ? : 

What is Edge Detection ? To mark the points on a image which are corresponds to Edge of an object. Normally it is the discontinuity in gradient of a pixel This points are indentified and than image is converted in to binary. (B&W)

Why Fuzzy Inferences System is used ? : 

Why Fuzzy Inferences System is used ? Different methods for Edge detection. Linear : Convolution mask, Sobel operator Non-linear : Laplacian-of-Gaussian, FIS Recurring Neural Network – most robust Fuzzy Inferences System Less computational effort Better result than Sobel method Specially in case of non uniform lighting

Overall System and Fuzzy Inferences System : 

Overall System and Fuzzy Inferences System Input Side Frame is taken out from a video Converted into grayscale image 4 non successive filters applied Sobel operator : horizontal derivatives Sobel operator : vertical derivatives High pass filter : Low pass filter : mean filter . . .

Overall System and Fuzzy Inferences System : 

Overall System and Fuzzy Inferences System FIS Input: All 4 pixel form 4 filtered images 3 membership function Rules Defuzzification : Centroid Output: Defuzzified value of pixel hence image FIS

Overall System and Fuzzy Inferences System : 

Overall System and Fuzzy Inferences System Output Side Resulting figure after FIS is taken Converted into binary image Captured as a frame Frames converted into video

Image : 1 : 

Image : 1 Origional Image

Image : 1 : 

Image : 1 FIS resulted Image

Image : 1 : 

Image : 1 Sobel oprator resulted Image

Image : 2 : 

Image : 2 Original Image Sobel Image FIS Image

Image : 3 : 

Image : 3 Original Image Sobel Image FIS Image

Image : 4 : 

Image : 4 Original Image Sobel Image FIS Image

Video : 

Video Origional Video Output video

Limitation : 

Limitation Slower than sobel method For well contrast image sobel method is more appropriate as it gives equally good result Tuning of rules is required for different kind of environment

Slide 17: 

Thank You

Slide 18: 

Thank You

Slide 19: 

Thank You

Slide 20: 

Thank You

Image : Filt : 

Image : Filt Back

Image : Membership Function : 

Image : Membership Function Back

All Filters : 

All Filters Back Sobel Horizontal Sobel Vertical High pass Mean Filter

Rules : 

Rules DH = L AND DV = L E = L DH = M AND DV = M E = H DH = H OR DV = H E = H DH = M AND HP = L E = H DV = M AND HP = L E = H DV = M AND M = L E = L DH = M AND M = L E = L DV = H AND DV(i+1) = H E= M DH = H AND DH(i+1) =H E= M DV = M AND DV(i+1) = H E= L DH = M AND DH(i+1)=H E = L DV(i+1) = L AND DV(i-1) = L AND DH(i+1) = L AND DH(i-1) = L  E = L Back Legends Inputs: DH : Sobel Horizontal DV : Sobel Verical HP : Hi pass M : Meanpass Output: E : Edge pixel Membership function : L : low M : Medium H : High