Project 4

Srikar Talluri 3036580613

Shoot the Pictures

I used two sides of my desk

Recover Homographies

Used Tool from project 3

Solved the following equation in the form Ax = b

Then added a singular 1 and then reshaped it into 3,3 matrix as desired

This gives us

Warp the Images

Simply following the procedure in the instructions:

I first made a list of grid points that would need to be mapped

I then took the transpose of this as I needed to multiply it with H. After getting the mapped points, I simply figured out a shift in the x and y direction that was needed to fit the image ensuring that all coordinates are positive. This was done by finding the minimum and simply shifting all coordinates by that amount.

I then mapped all coordinates to that shifted location. This gave the following result:

left warped to right image

Image Rectification

This was very simple as I constructed a new data_point_2 which contained rectangular coordinates of a square. In the correspondence tool, I selected the corners of my monitor. I then constructed an H_rect using the computeH function above and then finally ran warpImage(left, H_rect).

This resulted in the following image.

Blend the images into a mosaic

I followed the first approach of only warping one image and leaving the other as is. Since I already warped the left one, I decided to continue with this approach and leave the right as is.

After calculating the shift needed, from the warpImage function, I shifted the right image by the necessary amount.

This resulted in the following image:

Part B

Detecting corner features in an image

We run the harris selector and plot the points

Extracting a Feature Descriptor for each feature point

After implementing the ANMS feature selector, these are the points that resulted.

What have you learned?

Tell us what's the coolest thing you have learned from this project.

One of the things that I learned was how powerful simple algorithms can be in the context of image stitching and mosaics. I learned about Harris feature detection, blurring, and matching these feature descriptions between two images to blend them.

The coolest thing I learned from this project is definitely the ANMS method which was incredibly powerful in terms of mimicking human perception, like selecting significant points in an image that "stand out" based on context.