This paper addresses the problem of generating possible object locations
for use in object recognition. We introduce Selective Search
which combines the strength of both an exhaustive search and segmentation.
Like segmentation, we use the image structure to guide
our sampling process