Computer Vision General Topics - 2022

Homograrphy Transformation

Purpose:

1. Obtain a 3*3 matrix H  to transform the image to the desired position and angle. 

2. (Map my "flower" image to my "window" images shot from different angles.)

Procedures:

1. Obtain the coordinates of each image corner by using GIMP.

2. Determine the transformation matrix H by using the coordinates.

3. Transform each point through x′ = Hx.

4. Place the transformed values into the images.

Panorama Image

Purpose:

1. Create panorama by connecting images with different views.

Procedure:

1. Use SIFT to pair corresponding points between images.

2. Use RANSAC to identify inliers and outliers.

3. Calculate transformation matrix H.

4. Connect the images into a panoramic view.

Image Segmentation

Purpose:

1. Differentiate different objects in an image by using RGB segmentation and texture segmentation.

2. Expect that 

   1. the RGB segmentation works best with varied color channels.

   2. The texture segmentation works best with varied color intensities.

RGB Procedure:

1. Separate the images into Red, Green, and Blue channels.

2. Use Otsu algorithm to create a mask for each channel.

3. Combine individual masks as the final segmentation mask.

Texture Procedure:

1. Find spatial organization of colors or intensities.

2. Divide images into regions of interest and classify those parts.

Projective Stereo Reconstruction

Procedure:

• Estimate  homographies for image rectification

• Construct point correspondences

• Reproject your points into 3D coordinates