Question

I am working on a Computer Vision problem, in which I have to merge regions of an image. A region (or blob) is defined by its lines, i.e. the following region of O:

  0123456789
0 XXXXOXXXXX
1 XXXOOOXXXX
2 XXXOOXXXXX
3 XXXOXXXXXX
4 XXXXXXXXXX

is defined by :

row: 0, cols: 4-4
row: 1, cols: 3-5
row: 2, cols: 3-4
row: 3, cols: 3-3

I chose this datastructure because I need to be able to find quickly the neighbors of a region, i.e. all the pixels that “touch” it.

Now, my problem is that I want to merge two regions, i.e. compute the union of them. This means that I may end up with several ranges of columns in my data structure shown above.

With this setup, I have two questions:

1. In C, what is the best datastructure for this data ? A typical image is 16×16, which means not that many rows/columns. I will do a lot of merging (the goal of the thing is to start with one region per pixel and end with one big region, i.e. 16×16 – 1 merges). I could go with a pointer and allocate/free the things, or use a char* to store the cols and parse it afterwards, for instance.

2. How to efficiently merge two regions? I need to find potential columns that are adjacent to merge them (e.g., 3-5 and 6-9 become 3-9), preferably without always reallocating and copying things.

Answer 1

If your data is for a two color image…black and white, why not just use an array of unsigned shorts that is 16 elements long?

unsigned short image[16]

The merging could be done with bitwise logic which is pretty efficient over a 16 element array.