Homework 3: Image Resizing Using Seam Carving
In this assignment an algorithm for resizing an image with minimal noticeable distortion is implemented using seam carving: The following steps were taken following this paper by Shai Avidan & Ariel Shamir. 1. Computing the Energy function (imenergy.m) 2. Computing the optimal horizontal seam (horizontal_seam.m) 3. Removing the Seam (remove_horizontal_seam.m) 4. Resize (shrnk.m)
Results
Input Image: union-terrrace.jpg
This is the target image we want to resize.
Energy Function Output: 2a.jpg
The energy image shows when there is a large contrast from one area to the other (from dark to light or vise versa). For example, you can see that the energy image is very bright around the circumference of the yellow tables. This indicates a sharp change in intensity from the deck to the table. Our seems will most likely go around these areas because removing pixels there will generate visible artifacts in the resized image.
cumulative minimum energy array: 2b.jpg
The cumulative energy image shows the progression from left to right of the cumulative energy. This image has the same number of pixels as union-terrace.jpg. We start with low intensity values (blue) and as we move to the right they gradually get higher values(yellow/orange). In the last column the pixels with the lowest energy path leading to them will be bluish while those of which have a high energy path will appear “orangish”.
100 Columns Removed: 1a.jpg
Result of removing 100 columns from union-terrace.jpg
100 Rows Removed: 1b.jpg
Result of removing 100 rows from union-terrace.jpg
100 Columns and 100 Rows removed: 1c.jpg
Result of removing 100 columns and 100 rows from union-terrace.jpg
