opencv - Image Segmentation -

so trying write code lets me segment fuses see in picture below. have come 2 approaches:

1) based on color. threshold using opencv's inrange function. approach works fuses except brown fuse. brown fuse similar in colour fusebox , therefore it's hard segment out.

2) considered thresholding image heavily can detect white points/terminals on fuses using opencv simpleblobdetector. filter out blobs distances each other. know size of fuses, can filter out invalid fuses. approach works fuses white 1 appears in thresholded images.

i hoping pointer on how segment such image. background subtraction work?

my experience segmentation single approach not work difficult segmentation. if 1 algorithm works brown , other white, union of 2 should yield complete result. know nice have 1 elegant algorithm, many of best results have had resort hybrid of multiple techniques.

i'd consider separating channels rgb , hue, saturation, , value , looking @ each channel separately. browns similar in color have different saturation or color channel values. adding , subtracting different channels can enhance contrast. simple in many cases produces fast , simple output can used thresholding, watershed (see below), or perhaps background subtraction.

i think might want try watershed algorithm. many examples , explainations available. watershed requires provide mask contains background (the fusebox , table) , piece of each of foreground objects (fuses). understand it, can detect contacts on fuses piece done.

another approach accept can't see brown fuses. if can detect empty slots , every other color, may able know deduction brown ones are.

it's had know work beforehand without experimentation, should give ideas of how improve have.