HW1_MN

An Image Processing Project.

Timeline and Overview

The project was completed in 3 weeks and 2 days. Implementation ideas came to me successively.

Week 1: Nearest-Neighbor Interpolation

• Developed simple functions using nearest-neighbor interpolation.
• Used the round() function to find the nearest neighbor (also worked with floor()/ceil()).
• Added 1 to obtain a correct result.
• Implemented nn_resize:
  • Found valid scaling factors.
  • Used round() again for correct results.
• RGB functions were straightforward, based on the other functions.

Week 2: Bilinear Interpolation

• Spent 3-4 days correcting bilinear_resize and bilinear_rotate functions.
• The more permissive checker was helpful.
• Faced most obstacles in the optimization part.
• Additional 2-3 days were necessary for debugging.

Key Functions:

1. bilinear_2x2:

   • Used the relation presented in the PDF document.
   • Introduced valid values in corners.

2. bilinear_coef:

   • Similar to bilinear_2x2.
   • Swapped x and y (f(y,x)) due to matrix representation.

3. bilinear_resize and bilinear_rotate:

   • Experience from Week 1 was helpful.
   • Found interpolation coefficients by calculating I values at corresponding points.
   • Used the relation from the document to find the final value.
   • bilinear_rotate was similar to bilinear_resize, with an additional check for point existence in the image.

4. RGB functions were quickly implemented, based on other functions.

Week 3: Bicubic Interpolation

• Spent 2 more days debugging.

Key Functions:

1. precal_d:

   • Required much attention due to many conditions.
   • Used pen and paper for implementation.

2. Auxiliary functions (fx, fy, fxy):

   • Simple, relations were in the document.

3. bicubic_resize:

   • Took a long time to write.
   • Used parts of bilinear_resize for the first part.
   • Performed Matrix = Matrix' operation for correct element order:

     a b  T  a c
         -->       (To avoid errors)
     c d     b d
     

   • Used mod and floor to stay within 0-3 range (as per document's sum indices).
   • Saved interpolated value in matrix R at the end.

4. RGB function was straightforward, dependent on resize_bicubic.

Conclusion

This project was complex but interesting. The provided document and skeleton were helpful.