{[ promptMessage ]}

Bookmark it

{[ promptMessage ]}

# ps4 - Problem Set 4 Image Segmentation Due 3/19 For this...

This preview shows pages 1–3. Sign up to view the full content.

Problem Set 4: Image Segmentation Due 3/19 For this problem set, you are to turn in an overview document covering parts 1 and 2. The overview document should be composed as described in Overview Requirements . You should read this document carefully. We have provided an example overview document written as if for Problem Set 2 from CS 312, Spring 2008. Submission in PDF is preferred; however, you can also submit your overview document as a .doc or a .txt file. In particular, starting with this problem set, it's your job to convince us that your code works, by describing your implementation and by presenting a testing strategy, test cases, and results. Example: sigma = 0.5 , k = 500 , and minSize = 50 . Part 1: Graph-based Image Segmentation (55 points) The idea: The goal of image segmentation is to partition an image into a set of perceptually important regions (sets of contiguous pixels). It's difficult to say precisely what makes a region "perceptually important", and in different contexts it could mean different things. The algorithm you are going to implement uses the variation in intensity as a way to distinguish perceptually important regions (also called segments). This criteria seems to capture our intuitive notion of a segmentation quite well. To illustrate how the algorithm uses intensity variation as a way to segment an image, consider the grayscale image below. The image to its right was obtained by applying the image segmentation algorithm and then randomly coloring the segments. All of the pixels in the large rectangular region on the right have the same intensity, and so they are segmented together. The rectangular region on the left is considered its own segment because it is generally darker than the region on the right, and there is not much variation between the intensity of adjacent pixels. Interestingly, the noisy square on the right is considered a single region, because the intensity varies so much. There are some large differences between the intensities of adjacent pixels in the small square region; larger even than the difference between pixels at the border of the two large rectangular regions. This tells us that we need more information than just the difference in intensity between neighboring pixels in order to segment an image. Note that the algorithm found a few thin regions in the middle of the image. There is no good theoretical reason why these regions should be there: they were simply introduced as an artifact of smoothing the image, a process that will be discussed later. Parameters: sigma = 0.5 , k = 500 , minSize = 50 .

This preview has intentionally blurred sections. Sign up to view the full version.

View Full Document
Image segmentation provides you with a higher-level representation of the image. Often it is easier to analyze image segments than it is to analyze an image at the pixel level. For example, in the 2007 DARPA Urban Challenge, Cornell's vehicle detected road lines using this image segmentation algorithm. First, a camera in the front of the car takes a picture. Then the image is segmented using the algorithm you are
This is the end of the preview. Sign up to access the rest of the document.

{[ snackBarMessage ]}

### Page1 / 7

ps4 - Problem Set 4 Image Segmentation Due 3/19 For this...

This preview shows document pages 1 - 3. Sign up to view the full document.

View Full Document
Ask a homework question - tutors are online