Results & What’s Next

Results

After running the MATLAB code we wrote, we acquired the following results for the original sunflower image.

(Click on any image below)

This original image was 480 x 640 pixels.
This original image was 480 x 640 pixels.
Our original image split into the three RGB channels looks like this.
Our original image split into the three RGB channels looks like this.
The compression rate acquired for this image made us realize that we should be more considerate of our selection of k.
The compression rate acquired for this image made us realize that we should be more considerate of our selection of k.
95% Similarity
85% Similarity
75% Similarity
70% Similarity
65% Similarity
60% Similarity
50% Similarity
25% Similarity

We also plotted the compression ratio for each k value from 0 to  :

What’s Next

So our SVD implementation utilized a basic low-rank approximation algorithm. This doesn’t necessarily translate as the most optimized SVD method to compress an image. In fact, there exists other methods that can help us improve our ratio. In the future, we would like to look into those, implement them and compare it against basic low-rank approximation.