Due: 14.12

    - Download/choose an image.
    - Select a color space.
    - Select and implement quantization scheme.
    - Display color histogram and Quantized image.
        Note: If you encouter problem displaying 3D histogram, display a set of 2D views.
    - Try two different quantizations (number of  levels in each dimension (4, 8, 16, etc.) )
        and show the corresponding quantized  image.
    - Perform 1-2 small changes to the image (for example - rotation, noise, etc...)
        For each change display the resultant histograms. Is this a substantial change?
 

2. Localization/Segmentation
    - Choose 2 dominant colors (by looking at the histogram).
    - For each chosen color, mask out regions that can be represented by that color.
       Produce 2 binary images.
       (Note, you may want to introduce thresholds to ensure substantial-size regions, etc.)

3. Color metrics & Image Matching

    We have seen several color distance metrics in class. Choose two similarity measures from the list below
    (you may suggest one distance measure of your own) and compute the similarity between the query image, Q, and the target image, Ti
 

    Target1, Target2, Target3, Target4, Target5).
 

Mean color: 

Dominant color: 

Histogram Intersection:

- Rank the images according to each chosen similarity measure and summarize in a table.
- Discuss the results.

Note: d2 is the Euclidean distance.

4. Histogram Intersection (H.I.)
    - Prove: In case of #pixels (query/model) != #pixels (target/image) the H.I. (d(q,t)) as defined in class is not a metric.
    - Refine: Suggest a way to update the H.I. measure, such that it is a metric.