6.003 homework #11 solutions - mitweb.mit.edu/6.003/f11/www/handouts/hw11-solutions.pdf · 2011....

3
6.003 Homework #11 Solutions Engineering Design Problem 1. Image reconstruction The rows and/or columns of the following images have been blurred. Figure out a way to sharpen each image, and identify the building. Here are thumbnails of the images: a1 a2 b1 b2 c1 c2 The images are available in machine-readable form (buildings.zip) on the 6.003 website. In general, blurring could be caused by many different sorts of transformations. We can start by making the simplest 6.003 assumption: that the blurring results from a system with a single pole. Such a system would have a system function of the form H blur (z )= 1 - p 1 - pz -1 where |p| < 1 if the system is stable. Such a system would blur the image if it were a lowpass filter, i.e., 0 <p< 1. If the image were blurred by a system with a single pole, then it could be deblurred with a system with a single zero: H deblur (z )= 1 - pz -1 1 - p with the corresponding difference equation y[n]= x[n] - px[n - 1] 1 - p The blurring appears to be mostly horizontal in the first row of pictures. Thus we should try passing rows of pixels through the deblurring difference equation. We could process the

Upload: others

Post on 01-Oct-2020

4 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: 6.003 Homework #11 Solutions - MITweb.mit.edu/6.003/F11/www/handouts/hw11-solutions.pdf · 2011. 11. 24. · 6.003 Homework #11 Solutions Engineering Design Problem 1. Image reconstruction

6.003 Homework #11 Solutions

Engineering Design Problem1. Image reconstruction

The rows and/or columns of the following images have been blurred. Figure out a wayto sharpen each image, and identify the building. Here are thumbnails of the images:

a1 a2

b1 b2

c1 c2The images are available in machine-readable form (buildings.zip) on the 6.003 website.

In general, blurring could be caused by many different sorts of transformations. We canstart by making the simplest 6.003 assumption: that the blurring results from a systemwith a single pole. Such a system would have a system function of the form

Hblur(z) = 1− p

1− pz−1

where |p| < 1 if the system is stable. Such a system would blur the image if it were alowpass filter, i.e., 0 < p < 1.If the image were blurred by a system with a single pole, then it could be deblurred witha system with a single zero:

Hdeblur(z) = 1− pz−1

1− p

with the corresponding difference equation

y[n] = x[n]− px[n− 1]1− p

The blurring appears to be mostly horizontal in the first row of pictures. Thus we shouldtry passing rows of pixels through the deblurring difference equation. We could process the

Page 2: 6.003 Homework #11 Solutions - MITweb.mit.edu/6.003/F11/www/handouts/hw11-solutions.pdf · 2011. 11. 24. · 6.003 Homework #11 Solutions Engineering Design Problem 1. Image reconstruction

6.003 Homework #11 Solutions / Fall 2011 2pixels left to right (i.e., “causally”) or right to left (“anti-causally”). A similar procedurecould be used for columns of pixels in the middle row.Answers:The first row of pictures were blurred by passing each row of pixels through a system withthe pole at 0.985 processing from left to right. The resulting pictures can be unblurred bypassing each row through a system with a zero at 0.985 processing from left to right.The second row of pictures were blurred by passing each column of pixels through asystem with the pole at 0.985 processing from bottom to top. The resulting pictures canbe unblurred by passing each column through a system with a zero at 0.985 processingfrom bottom to top.The third row of pictures were blurred by passing each column of pixels through a systemwith the pole at 0.985 processing from top to bottom. The resulting pictures can beunblurred by passing each column through a system with a zero at 0.985 processing fromtop to bottom.Results after unblurring are shown below.

Page 3: 6.003 Homework #11 Solutions - MITweb.mit.edu/6.003/F11/www/handouts/hw11-solutions.pdf · 2011. 11. 24. · 6.003 Homework #11 Solutions Engineering Design Problem 1. Image reconstruction

6.003 Homework #11 Solutions / Fall 2011 3

a1: biology building a2: Building 39

b1: Medical Center b2: Building 13

c1: Building 9 c2: Hayden