The Wikipedia article is pretty good. As long as your grids are small, just about anything will work.
Earlier this fall I did some flood filling on 10 megapixel scanned images. (The problem was to remove black edges from book pages that had been scanned on a photocopier.) In that case there are only two colors so I essentially treated the problem like a search in an undirected graph, with each pixel connected to its neighbors along the four compass directions. I maintained a separate bitmap to keep track of which pixels had been visited.
The main findings were
Don't try recursive depth-first search. You really want an explicit data structure.
An auxiliary queue uses much less space than a stack. About forty times less space. In other words, prefer breadth-first search to depth-first search.
Again, these findings apply only to grids with multiple megapixels. On a nice small grid like the one shown in your question, any simple algorithm should work.
与恶龙缠斗过久,自身亦成为恶龙;凝视深渊过久,深渊将回以凝视…