Sub Pixel Blob Analysis

    Dear All,


    My name is Ryo Saito from LINX Corporation in Japan.

    I want to get the result of Blob Analysis with sub pixel but I have no idea. If you have any ideas, could you share it with me?


    Ryo Saito

    Based on our Blob-Analysis VA operator we produce several features per found object.

    For example is the bounding box pixel precise due to the binary input image.


    The center of gravity (COG) is not normalized and requires to be divided by the area to get the real X and Y coordinate.

    Based the precision of the used division and the amount of pixels per object you will get a sub-pixel precise result.

    COG_Xnorm = COG_X / Area

    COG_Ynorm = COG_Y / Area

    In case you calculate the COGnorm in software you can use a data-type like double to get a sub-pixel precise result.


    If you want to do this in VA you can use fractional bits to get sub-pixel precise coordinates.

    As an example you can shift COG by 10bit left and divide by the area.

    The sub-pixel precision is 210 = 1024. So in between of two integer coordinates 1024 sub-pixel steps represent the possible precision.