IDEC DATAVS2 Series - Manual (Page 78)

SVS2 Series Instruction Manual
73
16.5. ► Blob / Blob analysis
A blob (blob stands for Binary Large Object) is a set of adjacent pixels having the same brightness.
Typically, they are searched and identified in binarized images (►), in an attempt to group all bright
pixels and all dark pixels together into connected structures.
There are two key techniques to identify these structures and they focus on two different aspects:
Individual pixel: in this approach, once a (say) dark pixel is identified, all adjacent pixels are analysed
recursively and the dark pixels are attached to the blob. This process is applied to each and every
image pixel;
Line: the image is scanned line by line to identify dark and bright regions, which are appropriately
attached to the previously identified blobs as scanning proceeds.
Obviously enough, the second approach is faster, because the image is only processed once!
To further improve performance, filters may be set for the minimum number of pixels a connected
structure must have in order to be correctly recognised as a blob, as is the case (for instance) with
groups of pixels due to noise.
Filters may also be set for the maximum number of pixels in a blob: this solves the problems
associated with those clusters that touch on the perimeter of the inspection area.
Once the blobs have been identified, certain pieces of information may be obtained:
o
area;
o
bounding box (rectangle that encompasses the blob, useful for approximating the perimeter);
o
moments (to define orientation);
o
convex perimeter (provides a more accurate perimeter approximation than bounding box).
Undoubtedly, this analysis technique offers significant advantages, namely processing speed and the
ability to accept variations in the orientation and size of the target image.
However, identification proves difficult in the case of:
o
clusters having weak contrast to the background;
o
stacked objects or objects touching each other;
o
objects that closely resemble each other.