image - An alternative algorithm for blob detection -

image - An alternative algorithm for blob detection -

here's sample image of experiment processed finding info of blobs (centre, dimension etc.). image below contains noise, few bold lines, , of course of study few blobs.

my aim programme fpga using image processing bundle "sapera apf". sapera apf application programming teledyne dalsa frame grabbers perform image processing. , unfortunately doesn't allow write code in conventional way, neither allows utilize basic coding features (if else statements, loops etc.) supports inbuilt functions in form of code blocks, , can arrange these blocks. fyi, here's link list of functions supported sapera.

now, first objective test algorithm on matlab thereafter implementing on sapera. here approach in matlab:

applied mean filter (for step 3) thresholded image using adaptive thresholding (returning binary image) removed lines checking connectivity, lines connected above 100 pixels (for improve connectivity mean filter applied first) now since blobs remaining, location identified (with grayvalue=255) , corresponding pixels taken in original image.

this simple algorithm works fine in matlab. real problem step 3 (i.e. connectivity) of course of study had write loops going each pixel checking connectivity, can't done in sapera. looking other simpler algorithm separate blobs in images, without using loops , conditional statement, can implemented in sapera, need matlab code or algorithm.

as seems, sapera operations belong family of point-to-point or neighborhood operations (= image processing), , there little can help segment out blobs. (indeed, there no function homecoming "a little number of something"; homecoming total images or scalar values.)

as workaround can

limit processing part of involvement (a vertical stripe sure contain blobs - can ignore sides);

use horizontalsum function image profile in can find peaks corresponding blobs (you'll have programme yourself, amount of info much reduced);

similarly, utilize verticalsum find horizontal limits.

given first-class separation of blobs, should plenty delimit bounding boxes every blob.

as see on plot of vertical profile of image, peak detection easy (provided avoid lateral lines).

and horizontal profile seamless.

image algorithm matlab image-processing detection