In this component, you will write code to detect discriminating features in an image and find the best matching features in other images. Because features should be reasonably invariant to translation, rotation (plus illumination and scale if you do the extra credit), you'll use a feature descriptor discussed during lecture and you'll evaluate its performance on a suite of benchmark images. As part of the extra credit you'll have the option of creating your own feature descriptors. If there are enough entries we'll rank the performance of features that students in the class come up with, and compare them with the current state-of-the-art.

For the second part of the assignment, you will apply your features to automatically stitch images into a panorama.

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.

This component has three parts: feature detection, feature description, and feature matching.

In this step, you will identify points of interest in the image using the Harris corner detection method. The steps are as follows (see the lecture slides/readings for more details) For each point in the image, consider a window of pixels around that point. Compute the Harris matrix H for that point, defined as:

where the summation is over all pixels p in the window. The weights w_p should be chosen to be circularly symmetric (for rotation invariance). A common choice is to use a 3x3 or 5x5 Gaussian mask. Note that these weights were not discussed in the lecture slides, but you should use them for your computation. Predefined 5x5 and 7x7 Gaussian kernels are also available under features.h in the provided skeleton code.

Note that H is a 2x2 matrix. To find interest points, first compute the corner strength function (the "Harris operator").

Once you've computed c for every point in the image, choose points where c is above a threshold. You also want c to be a local maximum in at least a 3x3 neighborhood.

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. You'll create the descriptor using the following steps in the warpComputeFeatures function in the code skeleton (also see the lecture notes).

1. Prefilter the image using the 7x7 Gaussian filter. You may find the ImageLib "Convolve" function (in Convolve.h) useful here.
2. Determine the homography (rotation, translation and scale) required to warp an 8x8 image to a rotated 40x40 window centered at the feature.
3. Use the WarpGlobal function to acquire the values of the 8x8 downsampled image.
4. Update the feature descriptor with this data.

void WarpGlobal(CImageOf src, CImageOf& dst, CTransform3x3 M, WarpInterpolationMode interp);

WarpGlobal performs an inverse warp of the source image into the destination image. In other words, for every pixel of the destination image, the pixel in source image is computed using the specified transformation (and interpolated, if required). The transformation is specified by a 3x3 homography matrix that can represent rigid, affine, or perspective transformations.

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. This part of the feature detection and matching component is mainly designed to help you test out your feature descriptor. You will implement a more sophisticated feature matching mechanism in the second component when you do the actual image alignment for the panorama.

The simplest approach is the following: write a procedure that compares two features and outputs a distance between them. For example, you could simply sum the absolute value of differences between the descriptor elements. You could then use this distance to compute the best match between a feature in one image and the set of features in another image by finding the one with the smallest distance.

Two possible distance measures are:

1. Use a threshold on the match score. This is called the SSD distance, and is implemented for you as match type 1, in the function ssdMatchFeatures.
2. Compute (score of the best feature match)/(score of the second best feature match). This is called the "ratio test"; you must implement this distance.

Now you're ready to go! Using the UI and skeleton code that we provide, you can load in a set of images, view the detected features, and visualize the feature matches that your algorithm computes.

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.

You should also test the matching against the images you will take for your panorama (described in next component).

Follow these steps to get started quickly:

1. Download the skeleton code.
2. Download some image sets: graf, Yosemite. (Also included are some SIFT feature and database files.)
3. Download the solution EXE.

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

• Features
  with no command line options starts the GUI. Inside the GUI, you can load a query image and its corresponding feature file, as well as an image database file, and search the database for the image which best matches the query features. You can use the mouse buttons to select a subset of the features to use in the query.
   
• Features computeFeatures imagefile featurefile [featuretype]
  uses your feature detection routine to compute the features for imagefile, and writes them to featurefile. The param featuretype specifies which of your types of features (if you choose to implement another feature for extra credit) to compute.
   
• Features matchFeatures featurefile1 featurefile2 threshold matchfile [matchtype]
  takes in two sets of features, featurefile1 and featurefile2, and matches them using your matching routine (the matching routine to use is selected by [matchtype]; 1 (SSD) is the default). The threshold for determining which matches to keep is given by threshold. The results are written to a file, matchfile, which can later be read by the Panorama program.
  
  Until you write your feature matching routine, the features are matched by minimizing the SSD distance between feature vectors.
   
• Features matchSIFTFeatures featurefile1 featurefile2 threshold matchfile [matchtype]
  same as above, but uses SIFT features.
   
• Features roc featurefile1 featurefile2 homographyfile [matchtype] outputfile
  creates the points necessary for the ROC curve for the feature you implement. You will use these values to create plots for your ROC curves.
   
• Features rocSIFT featurefile1 featurefile2 homographyfile [matchtype] outputfile
  is the same as above, but uses the SIFT file format.

We have given you a number of classes and methods to help get you started. The only code you need to write is for your feature detection methods and your feature matching methods, all in features.cpp. Then, you should modify computeFeatures and matchFeatures in the file features.cpp to call the methods you have written. We have provided a function dummyComputeFeatures which shows how to create the code to detect and describe features, as well as how to integrate it into the system. The function ssdMatchFeatures implements a feature matcher which uses the SSD distance and which demonstrates how a matching function should be implemented. The function warpComputeFeatures is the main function you will complete, along with the helper functions computeHarrisValues and computeLocalMaxima. You will also implement the function ratioMatchFeatures for matching features using the ratio test.

You will also need to generate plots of the ROC curves for your feature detecting and matching code (using the 'roc' option of Features.exe), and for SIFT. For both the Yosemite test images (Yosemite1.jpg and Yosemite2.jpg), and the graf test images (img1.ppm and img2.ppm), create a plot with four curves, one using your features with the SSD distance, one using your features with the ratio test distance, and the other two using SIFT (with both the SSD and ratio test distances; these curves are provided to you in the zip files for Yosemite and graf provided above).

We have provided scripts for creating these plots using the 'gnuplot' tool. Gnuplot is installed on the lab machines, at 'C:\Program Files\gnuplot\binaries\wgnuplot.exe' ('gnuplot' is also available on 'attu', the CSE instructional machine). To generate a plot with gnuplot (using a gnuplot script 'script.txt', simply run 'C:\Program Files\gnuplot\binaries\wgnuplot.exe script.txt', and gnuplot will output an image containing the plot. The two scripts we provide are:

• plot.roc.txt
  plots the ROC curves for the SSD distance and the ratio test distance. These assume the two roc datafiles are called 'roc1.txt' (for the SSD distance), and 'roc2.txt' (for the ratio test distance). You will need to edit this script if your files are named differently. This script also assumes 'roc1.sift.txt' and 'roc2.sift.txt' are in the current directory (these files are provided in the zip files above). This script generates an image named 'plot.roc.png'. Again, to generate a plot with this script, simply enter 'C:\Program Files\gnuplot\binaries\wgnuplot.exe plot.roc.txt'.
   
• plot.threshold.txt
  plots the threshold on the x-axis and 'TP rate - FP rate' on the x-axis. The maximum of this function represents a point where the true positive rate is large relative to the false positive rate, and could be a good threshold to pick for the computeMatches step. This script generates an image named 'plot.threshold.png.'
   

Here is a list of suggestions for extending the program for extra credit. You are encouraged to come up with your own extensions as well. Feature detection and matching is an active area in computer vision - there are many interesting techniques you can attempt here based on techniques on the literature and on your own ideas.

• Make your feature more contrast invariant. This was discussed in lecture.
• Implement sub-pixel refinement of feature positions (MOPS paper).
• Implement adaptive non-maximum suppression (MOPS paper).
• Make your feature detector scale invariant.
• Implement a method that outperforms the SSD ratio test for deciding if a feature is a valid match.
• 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), wavelet indexing (approach from lecture), locality-sensitive hashing.
• Try implementing a better feature descriptor. You can define it however you want, but you should design it to be robust to changes in position, orientation, and illumination. You are welcome to use techniques described in lecture (e.g., detecting dominant orientations, using image pyramids), or come up with your own ideas For this extra credit you'll need to compare it with the other features using the following function:
  
  Features benchmark imagedir [featuretype matchtype]
  tests your feature finding and matching for all of the images in one of the four above sets. The param 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. This will be used for the extra credit if you choose to do that.