CSE 576 Computer Vision  Spring 2010

Project 2:  Feature Detection and Matching

Date released: Monday, April 19, 2010

Date due: Wednesday, May 5, 2010 11:59pm
Late policy: 5% off per day late till Friday, May 7, 2010

Download Indri's slides (results from her project)

Synopsis

In this project, you will write code to detect discriminating features in an image and find the best matching features in other images.  Your features should be reasonably invariant to translation, rotation, and illumination, and (to a lesser degree) scale, and you'll evaluate their performance on a suite of benchmark images.  Scale is less important because it's a lot harder - however, any descriptor that is invariant to the other factors will be slightly scale invariant as well. You're not expected to implement SIFT!

To help you visualize the results and debug your program, we provide a working user interface that displays detected features and best matches in other images.  We also provide sample feature files that were generated using SIFT, the current best of breed technique in the vision community, for comparison.

Description

The project has three parts:  feature detection, description, and matching.
The Harris operator and descriptors work with gray-scale images, so the input images will need to be converted. The skeleton code has the function ConvertToGrey in features.cpp to convert the images to grayscale.

Feature detection

In this step, you will identify points of interest in the image using the Harris corner detection method.  

For each point in the image, consider a 5 x5  window of pixels around that point.  Compute the Harris matrix M for that point, defined as

where the summation is over all pixels (xk,yk) in the window W centered at location (x,y).   For the weights w(xk,yk) you can use a 5x5 Gaussian mask 

I_x and I_y indicate the x and y-gradients of a pixel, which you can approximate using masks, such as the Sobel operator.

To find interest points, first compute the corner response R

(Try k = 0.05)

Once you've computed R for every point in the image, choose points where R is above a threshold (Try threshold = 0.001).  You also want it to be a local maximum in at least a 3x3 neighborhood.
To calculate the trace and determinant of M, you can find the formula in slide 20 of the Harris lecture.

Feature description

Now that you've identified points of interest, the next step is to come up with a descriptor for the feature centered at each interest point.  This descriptor will be the representation you'll use to compare features in different images to see if they match.

1. Simple descriptor
Use a small square window (say 5x5) as the feature descriptor.  This should be very easy to implement and should work well when the images you're comparing are related by a translation. You can test this simple descriptor on the images in the easy dataset (http://cat.middlebury.edu/stereo/data.html)
You might also want to try to normalize the brightness values by dividing by the length of the descriptor vector (square root of the sum of the squares).

2. Advanced descriptor
You can define it however you want, but you should design it to be robust to changes in position, orientation (i.e., rotation), and illumination.  You are welcome to use techniques described in lecture (e.g., detecting dominant orientations, using image pyramids, using a disc instead of a square window), or come up with your own ideas. This is the main challenge of the assignment.

Feature matching

Now that you've detected and described your features, the next step is to write code to match them, i.e., given a feature in one image, find the best matching feature in one or more other images.

The skeleton code provided finds the SSD between all feature descriptors in a pair of images. The code declares a match between each feature and it's best match (nearest neighbor) in the second image.

For each feature in the first image, use SSD to find the best match (or no match) in the second image. The idea here is to find a good threshold for deciding if a match exists. To do this, compute (score of the best feature match)/(score of the second best feature match), and threshold on that.

Testing

Now you're ready to go!  Using the UI and skeleton code that we provide, or your own matlab code, you can load in a set of images, view the detected features, and visualize the feature matches that your algorithm computes. Matlab users may want to scope out the C++ code for tips on comparing the features.

We are providing a set of benchmark images to be used to test the performance of your algorithm as a function of different types of controlled variation (i.e., rotation, scale, illumination, perspective, blurring).  For each of these images, we know the correct transformation and can therefore measure the accuracy of each of your feature matches.  This is done using a routine that we supply in the skeleton code.

Everybody

1. Download some image sets: leuven, bikes,  wall, graf
   Included with these images are

   • SIFT feature files, with extension .key
   • database files used by the skeleton code, .kdb
   • homography files, containing the correct transformation between each pair of images, named HXtoYp where X and Y are the indeces of the images. For example, the transformation between img2.ppm and img3.ppm is found in H2to3p.
2. Easier translation sequences can be found at http://cat.middlebury.edu/stereo/data.html. This dataset can be used to test your simple descriptor where the descriptor is invariant to translation.

C++ Users

• Get access to FLTK. (Here are local copies for windows and linux, but see below about using it on linux.) **disclaimer** the skeleton code was designed to be used with FLTK-1. It has not been tested with FLTK-2.
  On Windows you'll need to install it yourself. (If you unzip FLTK to somewhere other than  the directory above the project, you'll have to change the project settings to look for the include and library files in the correct location.) If you're using Linux on one of the CSE department machines, you don't need to download FLTK, since you can just use the libraries in /cse/courses/cse557/fltk-1.1.2 . The Makefile included in the zip expects to find fltk in /uns. If you already have it or are installing it in /usr/local, here are fltk installation instructions and an example of the Makefile for the skeleton code (you will need to change the path to suit your own paths).
  

• Download the C++ skeleton code:

• Visual Studio 2003 version (unix users are suggested to download this version)
  
• Visual Studio 2005 version
  

• To Do

  The skeleton classes that need to be edited are in file features.cpp.
  The starting points for editing the classes are in functions: computeFeatures and matchFeatures. computeFeatures is the function that performs feature detection and feature description. matchFeatures is the function that performs the feature matching. Currently computeFeatures is calling a dummy function called dummyComputeFeatures, and matchFeatures is calling a dummy function called dummyMatchFeatures.
  
  1.For feature detection, you will need to modify the computeHarris method. The skeleton code provides the function to conver tthe image to grayscale and to convolve the image with a Gaussian mask.
  2. For feature description, you will need to write two different extractDescriptor methods that implement the two different ways of representing your features, and modify computeFeatures to call the two different extractDescriptor methods.
  3. For feature matching, you will need to write  one  function for the MatchingFeatures function that implement the ratio  threshold matching when using SSD. .

Running the program

After compiling and linking the skeleton code, you will have an executable cse576. This can be run in several ways:

• cse576 with no command line options starts the GUI. 

          To use the GUI:

• Load a query image: File -> Load Query Image
• Load its corresponding feature file: File -> Load query features
• Load the feature database: File -> Load image database
• You can then search the database for the image which best matches the query features: Image -> Perform query 

                    You can also use the mouse buttons to select a subset of the features to use in the query, and then perform the query
            Until you write your feature matching routine, the features are matched by minimizing the Euclidean distance between feature vectors.
            The GUI is used to visualize the location of your feature points and the matching. You will need to compute your features first (using the computeFeatures argument in the command line) to compute              your Harris features and save them to a file. Then only you can load your feature file and visualize it in the GUI.

• cse576 computeFeatures imagefile featurefile [featuretype]
  uses your feature detection routine to compute the features for imagefile, and writes them to featurefile. featuretype specifies which of your (at least two) types of features to compute.

• cse576 testMatch featurefile1 featurefile2 homographyfile [matchtype]
  uses your feature matching routine to match the features in featurefile1 with the features in featurefile2. homographyfile contains the correct transformation between the points in the two images, specified by a 3-by-3 matrix. matchtype specifies your type of matching algorithm to use. Returns the average pixel error between the matched feature and the real transformed point.

• cse576 testSIFTMatch featurefile1 featurefile2 homographyfile [matchtype]
  is the same as above, but uses the SIFT file format.

• cse576 benchmark imagedir [featuretype matchtype]
  tests your feature finding and matching for all of the images in one of the four above sets. imagedir is the directory containing the image (and homography) files. This command will return the average pixel error when matching the first image in the set with each of the other five images.

Matlab Users

You are a little more on your own here, you will probably find it useful to look at the skeleton code. The key element of testing is to take two images, generate lists of feature descriptors for them, and match them. For each feature match, see how far off the actual transformation is. This routine applies the transformation matrix (similar to applyHomography in the C++ code).

function [xNew, yNew] = applyHomography(x, y, h)

    d = h(7).*x + h(8).*y + h(9);

    xNew = (h(1).*x + h(2).*y + h(3)) ./ d;
    yNew = (h(4).*x + h(5).*y + h(6)) ./ d

What to Turn In

Download the .doc template for the report here. You are free to use other text processing tools like latex etc, however make sure that you have the same sections in your report.
The grading guidelines for project 2 is here.

In addition to your source code and executable, turn in a report describing your approach and results.  In particular:

• Describe your feature descriptor in enough detail that some one could implement it from your write-up
• Explain why you made the major design choices that you did
• Report the performance on the provided benchmark image sets. Show the data for each of the pairs (average distance, or another metric, such as the percentage of matches which were less than some threshold apart (be sure to report what the threshold is!)). Results are expected to decline for higher image numbers; consider graphing this decline.
• Compare the performance of the simple window descriptor, your feature descriptor, and SIFT features (provided)
• Describe strengths and weaknesses
• Describe any extra credit items that you did (if applicable)

This report can be a Word document, or pdf document.
Email Indri (indria@cs.washington.edu) writeup and your source code by Wednesday, May 5, 2010 11:59pm. Zip up your report, source code and images into a file with your name as the name of the file , eg. JohnDoe.

Extensions

For those who would like to challenge themselves, here is a list of suggestions for extending the program for a small amount of extra credit. You are encouraged to come up with your own extensions as well!

• Use a fast search algorithm to speed up the matching process.  You can use code from the web or write your own (with extra credit proportional to effort).  Some possibilities in rough order of difficulty:  k-d trees (code available here),  locality-sensitive hashing.
• Make your feature detector scale invariant.
• Implement a method that outperforms the above ratio test for deciding if a feature is a valid match.

Note:  
This project is slightly harder than project 1, so start early! Also we do not expect your descriptor to beat the SIFT descriptor.

---

---