I am converting a grayscale signature image to thin image but not getting the desired output

after applying the im2bw(img,graythresh(img)) function i get the following image

but the image is not smooth enough after applying thining on the said image i get the following

so how to get rid of extra roots from the thin image and get smooth signature skeleton

Answer 1

[Anand]

It looks like you need to clean up the binary image a little before thinning.

This worked for me:

bw = im2bw(img,graythresh(img));

% remove some of the noisy edges in the mask.
bw_clean = imopen(bw, strel('disk',1));

% remove small foreground objects.
bw_clean = bwareaopen(bw_clean, 25);

% invert the image and thin it.
bw_thin = bwmorph(imcomplement(bw_clean), 'thin', Inf);

[Sohail]

Hi Anand thanks for your response. let me ask you about the following image

here in the image there are some white pixels inside black stroke of the signature can we remove this by adding black pixels in place of white pixels inside the stroke

[Walter Roberson]

imclose the object to fill the small holes.

[Sohail]

Hi Walter Roberson how would you approach to correctly thin the grayscale image (attached with question) to thin image without having extra roots

[Sohail]

Whats the function to remove extra roots around after thining function

[Walter Roberson]

I would not do that. "Extra roots" is not well defined and "correctly thin" is not well defined. When it comes to signature verification, little turns can be characteristic, as can be stroke width.

However, you might find it interesting to dilate the object, then filter out small blobs such as with bwareaopen or bwareafilt, and then skeletonize it.

[Sohail]

actually i am only interested in endpoints after thinning and the roots produces extra endpoints which increase total endpoints count and that is what i dont want

[Walter Roberson]

Have you considered bwmorph() 'spur' ?

[Image Analyst]

imfill() will fill completely surrounded holes and leave exterior boundaries unchanged. imclose() will fill interior holes and will smooth out exterior boundaries.

[Image Analyst]

Have you seen section 23.4.6.4 of vision bib? It has handwriting algorithms.