ECE 533

9
ECE 533 Greg Yoblin & Joseph Marino Blemish Removal Using Edge Detection and Gaussian Smoothening

description

ECE 533. Greg Yoblin & Joseph Marino. Blemish Removal Using Edge Detection and Gaussian Smoothening. What is this about?. Many photographs that are taken (portraits) have blemishes that people wish to remove. - PowerPoint PPT Presentation

Transcript of ECE 533

Page 1: ECE 533

ECE 533Greg Yoblin & Joseph Marino

Blemish Removal Using Edge Detection and Gaussian Smoothening

Page 2: ECE 533

What is this about?•Many photographs that are taken (portraits) have blemishes that people wish to remove.

•There are programs that allow for the removal of these blemishes, however the methods are tedious and very manual.

Transformation done with our program

Page 3: ECE 533

What we have Added1. The first major difference between our program and existing programs

is that our program can be run on regions to remove multiple blemishes at once (Avoids the “eraser”-like methodologies from Photoshop, etc)

2. Designed to work on color and BW images.

3. Produces “ideal” filling information based on surrounding area to leave a minimal demarcation in the image to indicate removals

4. An automated blemish removal process that will scan the image and determine what things to remove on its own (future)

Page 4: ECE 533

How Does it Work?1. Contrast Stretching

2. Perform Canny Edge detection on selected region (produces binary image)

3. Flood fill background to isolate “blemishes” (similar to watersheding)

4. Invert the binary Blemish-Mask and AND it with the original region

5. Fill in newly removed areas with the average background value from the region

6. Perform a simple (size adaptive) Gaussian Blurring on the region

7. Replace the region in the original image without blemishes

Page 5: ECE 533

What is Canny?

1. Smoothen the Image with a Gaussian Mask.

2. Calculate Gradients using a vertical Sobel Mask, and a horizontal Sobel Mask. Sum the values to get the final gradient.

3. Calculate Edge Direction using

4. Determine the maximum value perpendicular to the direction of the gradient.

5. Apply a hysteresis to eliminate edges that fall below the low, include those that are higher than the high, and decide about the ones in the middle.

(y/x)tan 1

Page 6: ECE 533

What is Canny?

Page 7: ECE 533

How Does Removal Work?

Blemishes Removed

Page 8: ECE 533

Results

Page 9: ECE 533

Results