If you want more advanced pattern recognition, this is not it. But here is a really basic concept that will handle all simple rectangular shapes, including your example.

• "squish" all consecutive identical rows into only one row.
• Do the same for columns

In code, it looks like this:

my $IMAGE = <<END:
0000000000000000000000
0000000001111100000000
0000000001111100000000
0000000001111100000000
0000000001111100000000
0000011111111111110000
0000011111111111110000
0000011111111111110000
0000011111111111110000
0000000001111100000000
0000000001111100000000
0000000001111100000000
0000000001111100000000
0000000000000000000000
END

sub transpose {
    my @matrix = map [ split // ], split /\n/, shift;
    my $dim    = $#{ $matrix[0] };

    @matrix = map { my $x=$_; [ map { $matrix[$_][$x] } 0..$#matrix ] 
+}
              0..$dim;
    return join "", map { join("", @$_) . "\n" } @matrix;
}

sub squish {
    my $str = shift;
    $str =~ s/^(.*\n)\1+/$1/gm;
    return $str;
}

print transpose squish transpose squish $IMAGE;

__OUTPUT__
00000
00100
01110
00100
00000
[download]

I just threw this together, you could certainly make it more elegant, and resilient to the newlines in the data. But hopefully you get the idea. I chose to represent the image matrix as just a flat string because it makes squish simpler, but transpose would be a little more elegant with a 2-dimensional array instead of a string.

To match the exact output you wanted, you would also have to trim all leading and trailing trailing all-zero rows & columns.