Hi,

Maybe you can perform an analisys over the "Shape signature" like here http://basic-eng.blogspot.com.es/2006/01/signature-of-binary-objects.html

Once you have the signature, you can characterize each object (signature) and identify it by training a classifier as ANN or something like this.

If there are many features, they can be reduced using a PCA

Good luck

You can try convex polygon

Approach 1:

Find the difference between binary image and its convex hull & apply morphology to remove small regions & count the remaining no. of indents or foldings

____________________________________________________

%% Matlab program to check indents or foldings in binary image

clc;close all; clear all; warning off;

I=imread('indent.png');

BW=im2bw(I);

BW=imclearborder(BW);

figure,imshowpair(I,BW,'montage');

BW_convhull=bwconvhull(BW,'objects');

indent=xor(BW,BW_convhull);

figure,imshowpair(BW_convhull,indent,'montage')

indent_final=bwareaopen(indent,100); % remove small regions

figure,imshowpair(I,indent_final,'montage');

L=labelmatrix(bwconncomp(BW_convhull));

s=regionprops(L,'BoundingBox');

L1=bwlabel(imcrop(indent_final,s(1).BoundingBox));

L2=bwlabel(imcrop(indent_final,s(2).BoundingBox));

L3=bwlabel(imcrop(indent_final,s(3).BoundingBox));

disp('No. of indents in three objects respectively are:')

disp([max(L1(:)),max(L2(:)),max(L3(:))]);

_____________________________________________________

Result

No. of indents in three objects respectively are:

2 1 1

Approach 2:

Check http://syntacticsalt.com/2013/01/11/finding-dents-in-an-blobby-shape/

Found some useful blogs for you - http://mathminers.com/2016/05/27/image-analysis-using-tensor-flow/

http://mathminers.com/2016/05/18/face-recognition-analysis-using-python/